Phase 7: audit memory

- test_deep_agent: update patch target get_llm -> get_agent_llm (8 tests)
- test_device_ws: remove 5 tests for deleted agent_data_queue API
- test_schemas_v3: remove agent_run/agent_data/agent_complete from v2 compat list
- Delete test_agent_runner.py (superseded by test_agent_runner_v2.py)
- Delete test_agent_setup.py (superseded by test_journey_v2.py)
- Delete test_classify_file.py (_classify_file removed in v2 rewrite)

.env.example

@@ -2,7 +2,7 @@
 ENV=dev
 
 # ── Database ──────────────────────────────────────────────────────────────────
-DATABASE_URL=postgresql+asyncpg://postgres:postgres@localhost:5432/adiuva
+DATABASE_URL=postgresql+asyncpg://postgres:postgres@localhost:5432/adiuvai
 
 # ── Auth ──────────────────────────────────────────────────────────────────────
 JWT_SECRET=replace-with-a-long-random-secret
@@ -13,31 +13,72 @@ JWT_REFRESH_TOKEN_EXPIRE_DAYS=30
 # ── LLM ───────────────────────────────────────────────────────────────────────
 # LiteLLM model identifiers — change to swap providers without code changes.
 # Examples: gpt-4o, anthropic/claude-sonnet-4-20250514, gemini/gemini-pro, ollama/llama3
+#
+# API keys — only the key(s) matching your chosen provider(s) are required.
+# The correct key is picked automatically from the model prefix (e.g.
+# "anthropic/..." → ANTHROPIC_API_KEY, "gemini/..." → GOOGLE_API_KEY).
 OPENAI_API_KEY=
 ANTHROPIC_API_KEY=
 GOOGLE_API_KEY=
-LLM_MODEL=gpt-4o
+CEREBRAS_API_KEY=
-LLM_ROUTER_MODEL=gpt-4o-mini
+
+# Default model used by any agent that does not have a specific override below.
+LLM_MODEL=gpt-5-mini
+LLM_EMBED_MODEL=text-embedding-3-small
+
+# GitHub Copilot — leave empty to use the LiteLLM default token directory.
+# In Docker, point this to a named-volume path so tokens survive restarts.
+# GITHUB_COPILOT_TOKEN_DIR=
+
+# ── Per-agent model overrides ─────────────────────────────────────────────────
+# Leave a value empty to fall back to LLM_MODEL.
+# Each agent resolves its API key from the model prefix automatically.
+#
+# Intent classifier — routes user messages to the right domain agent.
+# A small/fast model (e.g. gpt-4o-mini) is usually sufficient here.
+LLM_MODEL_CLASSIFIER=
+
+# Home-agent — handles chat from the home screen (all tools available).
+LLM_MODEL_HOME_AGENT=
+
+# Floating-agent — handles contextual chat triggered from a task/project/note.
+LLM_MODEL_FLOATING_AGENT=
+
+# Unified-processor — processes local directory files (local agent runner).
+LLM_MODEL_UNIFIED_PROCESSOR=
+
+# Cloud-processor — fetches and processes data from cloud connectors.
+LLM_MODEL_CLOUD_PROCESSOR=
+
+# Setup-agent — guided journey to build an AgentConfig via WebSocket chat.
+LLM_MODEL_SETUP_AGENT=
+
+# Memory-extractor — Mem0-style extract/decide pipeline (Phase 2).
+# Defaults to gpt-4o-mini when empty (fast + cheap, temperature=0).
+LLM_MODEL_MEMORY_EXTRACTOR=
+
+# Memory-miner — proactive pattern mining from episodic history (Phase 5, Power+ only).
+# Defaults to gpt-4o-mini when empty.
+LLM_MODEL_MEMORY_MINER=
+
+# Memory-auditor — weekly contradiction scan + relation label canonicalization (Phase 7).
+# Defaults to LLM_MODEL when empty (a reasoning-capable model is recommended).
+LLM_MODEL_MEMORY_AUDITOR=
+
+# Scheduler — set to false to disable memory cron jobs (automatically false in tests).
+SCHEDULER_ENABLED=true
 
 # ── Stripe (leave empty to stub billing) ──────────────────────────────────────
 STRIPE_SECRET_KEY=
 STRIPE_WEBHOOK_SECRET=
 
-# ── AWS / S3 ──────────────────────────────────────────────────────────────────
-S3_BUCKET=adiuva
-S3_REGION=us-east-1
-S3_ENDPOINT_URL=
-AWS_ACCESS_KEY_ID=
-AWS_SECRET_ACCESS_KEY=
-# For MinIO (homelab): S3_ENDPOINT_URL=http://minio:9000
 
-# ── Vector Store ──────────────────────────────────────────────────────────────
+# ── Langfuse (leave empty to disable observability) ───────────────────────────
-# Pinecone is used when PINECONE_API_KEY is set; otherwise falls back to Qdrant.
+LANGFUSE_SECRET_KEY=
-PINECONE_API_KEY=
+LANGFUSE_PUBLIC_KEY=
-PINECONE_INDEX=adiuva
+# LANGFUSE_BASE_URL=https://cloud.langfuse.com        # EU (default)
-QDRANT_URL=
+# LANGFUSE_BASE_URL=https://us.cloud.langfuse.com     # US
-QDRANT_API_KEY=
+# LANGFUSE_BASE_URL=http://localhost:3000             # Self-hosted
-# For local Qdrant (homelab): QDRANT_URL=http://qdrant:6333
 
 # ── CORS ──────────────────────────────────────────────────────────────────────
 # Comma-separated list parsed by Settings (override default if needed)

.gitea/workflows/deploy.yaml

@@ -48,23 +48,23 @@ jobs:
           key: ${{ secrets.SSH_KEY }}
           script: |
             set -e
-            DEPLOY_DIR="/opt/adiuva-api"
+            DEPLOY_DIR="/opt/adiuvai-api"
             REPO_URL="http://10.0.0.119:3000/${{ gitea.repository }}.git"
             TAG="${{ gitea.ref_name }}"
 
             # ── Pull latest code ──
-            cd /tmp && rm -rf adiuva-api-deploy
+            cd /tmp && rm -rf adiuvai-api-deploy
-            git clone --depth 1 --branch "${TAG}" "${REPO_URL}" adiuva-api-deploy
+            git clone --depth 1 --branch "${TAG}" "${REPO_URL}" adiuvai-api-deploy
 
             # ── Sync source (preserve .env) ──
-            cp -rf /tmp/adiuva-api-deploy/app/ \
+            cp -rf /tmp/adiuvai-api-deploy/app/ \
-                   /tmp/adiuva-api-deploy/alembic/ \
+                   /tmp/adiuvai-api-deploy/alembic/ \
-                   /tmp/adiuva-api-deploy/alembic.ini \
+                   /tmp/adiuvai-api-deploy/alembic.ini \
-                   /tmp/adiuva-api-deploy/Dockerfile \
+                   /tmp/adiuvai-api-deploy/Dockerfile \
-                   /tmp/adiuva-api-deploy/docker-compose.yml \
+                   /tmp/adiuvai-api-deploy/docker-compose.yml \
-                   /tmp/adiuva-api-deploy/requirements.txt \
+                   /tmp/adiuvai-api-deploy/requirements.txt \
                    "$DEPLOY_DIR/"
-            rm -rf /tmp/adiuva-api-deploy
+            rm -rf /tmp/adiuvai-api-deploy
 
             # ── Verify .env ──
             if [ ! -f "$DEPLOY_DIR/.env" ]; then

.github/workflows/ci.yml

@@ -58,7 +58,7 @@ jobs:
       - uses: actions/checkout@v4
 
       - name: Build image
-        run: docker build -t adiuva-api:ci .
+        run: docker build -t adiuvai-api:ci .
 
       - name: Verify gunicorn installed
-        run: docker run --rm adiuva-api:ci gunicorn --version
+        run: docker run --rm adiuvai-api:ci gunicorn --version

.gitignore

@@ -21,6 +21,7 @@ env/
 .pytest_cache/
 htmlcov/
 .coverage
+tests/fixtures/private*/
 
 # Docker
 *.log

README.md

@@ -1,793 +0,0 @@
-# Adiuva Cloud API
-
-**AI-powered project management backend with E2E encrypted cloud storage, LLM orchestration, and a plugin marketplace.**
-
-Built with FastAPI · Python 3.12 · PostgreSQL · LangChain · Stripe · AWS S3
-
----
-
-## Table of Contents
-
-- [Overview](#overview)
-- [Architecture](#architecture)
-- [Key Features](#key-features)
-- [Tech Stack](#tech-stack)
-- [Getting Started](#getting-started)
-- [Docker Deployment](#docker-deployment)
-- [Environment Variables](#environment-variables)
-- [API Reference](#api-reference)
-- [Data Model](#data-model)
-- [AI Agent System](#ai-agent-system)
-- [Orchestration & Execution Plans](#orchestration--execution-plans)
-- [Middleware](#middleware)
-- [Storage Layer](#storage-layer)
-- [Billing & Tiers](#billing--tiers)
-- [Plugin Marketplace](#plugin-marketplace)
-- [Testing](#testing)
-- [Project Structure](#project-structure)
-- [License](#license)
-
----
-
-## Overview
-
-Adiuva Cloud API is the FastAPI backend that powers the **Adiuva Electron desktop app**. It provides LLM-powered chat orchestration, end-to-end encrypted cloud storage, a vector search engine, an encrypted backup system, a plugin marketplace with revenue sharing, and Stripe-based subscription billing across four tiers.
-
-### Design Principles
-
-1. **Never persist user data in plaintext** — the database stores only auth, billing, storage metadata, and marketplace data. All user content is E2E encrypted by the client before reaching the server.
-2. **Never expose prompts** — system prompts stay server-side; responses are sanitized to strip any leaked prompt fragments.
-3. **Never decrypt user blobs** — the backend performs only checksum verification; no decryption keys ever reach the server.
-4. **Stateless request handling** — all context comes from the client and JWT; no server-side session state.
-5. **Tier gates enforced server-side** — the server always reads the current tier from the database, never trusting client-reported values.
-
----
-
-## Architecture
-
-```
-┌──────────────┐      ┌────────────────────────────────────────────────────────┐
-│  Electron    │      │  FastAPI  (Uvicorn / Gunicorn)                         │
-│  Desktop App │────▶│                                                        │
-│  (Client)    │◀────│  Middleware: RateLimit → Sanitizer → CORS → Router     │
-└──────────────┘      │                                                        │
-                      │  ┌──────────────────┐  ┌────────────────────────────┐  │
-                      │  │  Auth Routes     │  │  Chat Routes               │  │
-                      │  │  Billing Routes  │  │    ↓                       │  │
-                      │  │  Storage Routes  │  │  Orchestrator (GPT-4o-mini)│  │
-                      │  │  Backup Routes   │  │    ↓ classify intent       │  │
-                      │  │  Plugin Routes   │  │  Agent Registry            │  │
-                      │  │  Vector Routes   │  │    ↓                       │  │
-                      │  │  Plans Routes    │  │  TaskAgent  | ProjectAgent │  │
-                      │  └──────────────────┘  │  NoteAgent  | CheckptAgent │  │
-                      │                        │  (GPT-4o + LangChain)      │  │
-                      │                        └────────────────────────────┘  │
-                      └────────────────────────────────────────────────────────┘
-                               │              │              │
-                      ┌────────▼───┐  ┌───────▼───────┐  ┌──▼─────────────┐
-                      │ PostgreSQL │  │  AWS S3       │  │ Pinecone /     │
-                      │ (Auth,     │  │  (E2E blobs,  │  │ Qdrant         │
-                      │  Billing,  │  │   backups)    │  │ (Vectors)      │
-                      │  Metadata) │  └───────────────┘  └────────────────┘
-                      └────────────┘
-                               │
-                      ┌────────▼───┐
-                      │  Stripe    │
-                      │  (Billing, │
-                      │   Connect) │
-                      └────────────┘
-```
-
----
-
-## Key Features
-
-1. **LLM-powered orchestration** — GPT-4o-mini classifies user intent and routes to the appropriate domain agent.
-2. **4 specialized AI agents** — Tasks (8 tools), Projects (6 tools), Timelines (4 tools), Notes (5 tools), all powered by GPT-4o via LangChain.
-3. **Execution plans & playbooks** — Server-side prompt template registry; clients receive only opaque template IDs, never raw prompts.
-4. **E2E encrypted cloud storage** — The backend never decrypts user data; SHA-256 checksum verification uses constant-time comparison to prevent timing attacks.
-5. **Cloud vector store** — Pinecone or Qdrant with user-isolated namespaces and encrypted blob payloads.
-6. **Encrypted backup system** — Tiered storage limits with `If-Modified-Since` support for efficient syncing.
-7. **Plugin marketplace** — Catalog, admin review/approval workflow, security checklist, and 70/30 revenue sharing via Stripe Connect.
-8. **Stripe billing** — Four-tier subscription model (Free / Pro / Power / Team) with checkout sessions and full webhook lifecycle handling.
-9. **JWT authentication** — Access + refresh tokens with bcrypt password hashing, SHA-256 token hashing, and automatic rotation.
-10. **Prompt IP protection** — Sanitizer middleware strips system prompts, reasoning markers, tool schemas, and agent routing metadata from all chat responses.
-11. **Tier-based rate limiting** — Sliding-window per-user limiter scaling from 20 to 200 requests/min by subscription tier.
-12. **Zero-trust data model** — User content is never stored in plaintext; the database holds only authentication, billing, and metadata records.
-13. **WebSocket streaming** — Real-time chat with 30-second heartbeat keep-alive and chunked text delivery.
-14. **Alembic migrations** — Versioned schema management with seed data for the plugin marketplace.
-15. **Comprehensive test suite** — In-memory SQLite + moto S3 mocks, per-tier test fixtures, and full API coverage without external dependencies.
-
----
-
-## Tech Stack
-
-| Package | Version | Purpose |
-|---|---|---|
-| `fastapi` | ≥ 0.115.0 | Web framework |
-| `uvicorn[standard]` | ≥ 0.34.0 | ASGI development server |
-| `gunicorn` | ≥ 22.0.0 | Production process manager |
-| `langchain` | ≥ 0.3.0 | LLM orchestration framework |
-| `langchain-openai` | ≥ 0.3.0 | OpenAI LLM provider integration |
-| `litellm` | ≥ 1.50.0 | Universal LLM gateway (100+ providers) |
-| `pydantic` | ≥ 2.10.0 | Data validation and serialization |
-| `pydantic-settings` | ≥ 2.7.0 | Environment-based configuration |
-| `python-jose[cryptography]` | ≥ 3.3.0 | JWT encoding and decoding |
-| `stripe` | ≥ 11.0.0 | Billing and payment integration |
-| `boto3` | ≥ 1.35.0 | AWS S3 client |
-| `slowapi` | ≥ 0.1.9 | Rate limiting utilities |
-| `sqlalchemy` | ≥ 2.0.0 | Async ORM and query builder |
-| `asyncpg` | ≥ 0.30.0 | PostgreSQL async driver |
-| `alembic` | ≥ 1.14.0 | Database migration management |
-| `bcrypt` | ≥ 4.2.0 | Password hashing |
-| `python-dotenv` | ≥ 1.0.0 | `.env` file loading |
-| `httpx` | ≥ 0.28.0 | Async HTTP client (used in tests) |
-| `websockets` | ≥ 14.0 | WebSocket protocol support |
-| `psycopg2-binary` | ≥ 2.9.0 | Synchronous PostgreSQL driver (Alembic) |
-| `pinecone` | ≥ 5.0.0 | Pinecone vector store client |
-| `qdrant-client` | ≥ 1.7.0 | Qdrant vector store client |
-| `pytest` | ≥ 8.0.0 | Test framework |
-| `pytest-asyncio` | ≥ 0.24.0 | Async test support |
-| `aiosqlite` | ≥ 0.20.0 | In-memory SQLite for tests |
-| `moto[s3]` | ≥ 5.0.0 | AWS S3 mock for tests |
-| `ruff` | ≥ 0.8.0 | Linter and formatter |
-
----
-
-## Getting Started
-
-### Prerequisites
-
-- Python 3.12+
-- PostgreSQL 16+
-- An OpenAI API key (for LLM features)
-- Stripe API keys (optional — billing stubs gracefully when unconfigured)
-- AWS credentials (optional — needed for S3 storage in production)
-
-### Installation
-
-```bash
-# Clone the repository
-git clone <repo-url> && cd adiuva-api
-
-# Create a virtual environment
-python -m venv .venv && source .venv/bin/activate
-
-# Install dependencies
-pip install -r requirements.txt
-
-# Configure environment
-cp .env.example .env
-# Edit .env with your DATABASE_URL, OPENAI_API_KEY, etc.
-```
-
-### Database Setup
-
-```bash
-# Start PostgreSQL (or use the Docker Compose database)
-docker compose up db -d
-
-# Run migrations
-alembic upgrade head
-```
-
-### Run the Development Server
-
-```bash
-uvicorn app.main:app --reload --host 0.0.0.0 --port 8000
-```
-
-Interactive API docs are available at [http://localhost:8000/docs](http://localhost:8000/docs) in development mode (`ENV=dev`). The `/docs` endpoint is disabled in production.
-
----
-
-## Docker Deployment
-
-### Quick Start
-
-```bash
-docker compose up --build
-```
-
-This starts two services:
-
-- **app** — FastAPI server on port `8000`
-- **db** — PostgreSQL 16 (Alpine) on port `5432` with a persistent volume and health checks
-
-The compose file also includes optional services for fully local deployments:
-
-- **minio** — S3-compatible object storage on ports `9000` (API) and `9001` (console)
-- **qdrant** — Vector search engine on ports `6333` (HTTP) and `6334` (gRPC)
-
-### Dockerfile Details
-
-The Dockerfile uses a multi-stage build:
-
-1. **Builder stage** — Installs Python dependencies into a virtual environment.
-2. **Runtime stage** — Copies only the venv, app source, and Alembic migrations. Runs as a non-root user (`appuser`).
-3. **Production server** — Gunicorn with 4 Uvicorn workers, 120-second timeout, listening on port 8000.
-
-```bash
-# Production command (run by the container)
-gunicorn app.main:app -k uvicorn.workers.UvicornWorker -w 4 --timeout 120 -b 0.0.0.0:8000
-```
-
----
-
-## Homelab / Self-Hosted Deployment
-
-You can run the entire stack locally on a homelab with **no cloud dependencies except the LLM provider**. The compose file includes MinIO (S3 replacement) and Qdrant (vector store) out of the box.
-
-### 1. Start all services
-
-```bash
-docker compose up -d
-```
-
-This starts PostgreSQL, MinIO, and Qdrant alongside the app.
-
-### 2. Create the MinIO bucket
-
-Open the MinIO console at [http://localhost:9001](http://localhost:9001) (login: `minioadmin` / `minioadmin`) and create a bucket named `adiuva`, or use the CLI:
-
-```bash
-docker compose exec minio mc alias set local http://localhost:9000 minioadmin minioadmin
-docker compose exec minio mc mb local/adiuva
-```
-
-### 3. Configure your `.env`
-
-```bash
-# Database (uses the compose PostgreSQL)
-DATABASE_URL=postgresql+asyncpg://postgres:postgres@db:5432/adiuva
-
-# S3 → MinIO
-S3_BUCKET=adiuva
-S3_REGION=us-east-1
-S3_ENDPOINT_URL=http://minio:9000
-AWS_ACCESS_KEY_ID=minioadmin
-AWS_SECRET_ACCESS_KEY=minioadmin
-
-# Vector store → local Qdrant (leave PINECONE_API_KEY empty)
-QDRANT_URL=http://qdrant:6333
-QDRANT_API_KEY=
-PINECONE_API_KEY=
-
-# Billing — leave empty to stub (no Stripe needed)
-STRIPE_SECRET_KEY=
-STRIPE_WEBHOOK_SECRET=
-
-# LLM — the only external service
-OPENAI_API_KEY=sk-...
-LLM_MODEL=gpt-4o
-LLM_ROUTER_MODEL=gpt-4o-mini
-
-# Auth
-JWT_SECRET=your-secret-here
-ENV=dev
-```
-
-### 4. Run migrations
-
-```bash
-docker compose exec app alembic upgrade head
-```
-
-### What runs where
-
-| Service | Runs on | Port | Notes |
-|---|---|---|---|
-| FastAPI app | Docker | 8000 | API server |
-| PostgreSQL | Docker | 5432 | Auth, billing, metadata |
-| MinIO | Docker | 9000 / 9001 | S3-compatible blob & backup storage |
-| Qdrant | Docker | 6333 / 6334 | Vector search (replaces Pinecone) |
-| Stripe | — | — | Stubbed when keys are empty |
-| OpenAI / LLM | Cloud | — | Only external dependency |
-
-> **Want fully offline AI too?** Set `LLM_MODEL=ollama/llama3` and `LLM_ROUTER_MODEL=ollama/llama3`, then add an Ollama container or point at a local Ollama instance. See the [LLM provider switching](#switching-llm-providers) section.
-
----
-
-## Environment Variables
-
-All variables are loaded from a `.env` file via Pydantic Settings. Source: `app/config/settings.py`
-
-| Variable | Type | Default | Description |
-|---|---|---|---|
-| `DATABASE_URL` | `str` | `postgresql+asyncpg://postgres:postgres@localhost:5432/adiuva` | Async SQLAlchemy connection string |
-| `JWT_SECRET` | `str` | `change-me-in-production` | HMAC secret for JWT signing |
-| `JWT_ALGORITHM` | `str` | `HS256` | JWT signing algorithm |
-| `JWT_ACCESS_TOKEN_EXPIRE_MINUTES` | `int` | `30` | Access token time-to-live |
-| `JWT_REFRESH_TOKEN_EXPIRE_DAYS` | `int` | `30` | Refresh token time-to-live |
-| `STRIPE_SECRET_KEY` | `str` | `""` | Stripe API key (empty = stub mode) |
-| `STRIPE_WEBHOOK_SECRET` | `str` | `""` | Stripe webhook signature secret |
-| `S3_BUCKET` | `str` | `""` | S3 bucket for encrypted blobs and backups |
-| `S3_REGION` | `str` | `us-east-1` | AWS region |
-| `S3_ENDPOINT_URL` | `str` | `""` | Custom S3 endpoint (e.g. `http://minio:9000` for MinIO). Leave empty for AWS. |
-| `AWS_ACCESS_KEY_ID` | `str` | `""` | AWS credentials |
-| `AWS_SECRET_ACCESS_KEY` | `str` | `""` | AWS credentials |
-| `PINECONE_API_KEY` | `str` | `""` | Pinecone API key (if set, Pinecone is used for vectors) |
-| `PINECONE_INDEX` | `str` | `adiuva` | Pinecone index name |
-| `QDRANT_URL` | `str` | `""` | Qdrant URL (used when Pinecone is not configured) |
-| `QDRANT_API_KEY` | `str` | `""` | Qdrant API key |
-| `OPENAI_API_KEY` | `str` | `""` | OpenAI key for LLM agent calls |
-| `LLM_MODEL` | `str` | `gpt-4o` | LiteLLM model identifier for agents (e.g. `anthropic/claude-3.5-sonnet`, `gemini/gemini-pro`, `ollama/llama3`) |
-| `LLM_ROUTER_MODEL` | `str` | `gpt-4o-mini` | Lighter model used for intent classification / routing |
-| `CORS_ORIGINS` | `list[str]` | `["app://.", "http://localhost:3000", "http://localhost:5173"]` | Allowed CORS origins |
-| `ENV` | `Literal` | `dev` | `dev` or `prod` — controls `/docs` visibility and SQL echo |
-
----
-
-## API Reference
-
-All routes are prefixed with `/api/v1`. **27 endpoints** total (25 REST + 1 WebSocket + 1 health check).
-
-### Health
-
-| Method | Path | Auth | Description |
-|---|---|---|---|
-| `GET` | `/api/v1/health` | No | Returns `{"status": "ok", "version": "0.1.0"}` |
-
-### Auth
-
-| Method | Path | Auth | Description |
-|---|---|---|---|
-| `POST` | `/api/v1/auth/register` | No | Create account with bcrypt-hashed password, returns `AuthTokens` |
-| `POST` | `/api/v1/auth/login` | No | Validate credentials, returns `AuthTokens` |
-| `POST` | `/api/v1/auth/refresh` | No | Rotate refresh token, returns new `AuthTokens` |
-| `GET` | `/api/v1/auth/me` | JWT | Returns `UserProfile` for the authenticated user |
-
-### Chat
-
-| Method | Path | Auth | Description |
-|---|---|---|---|
-| `POST` | `/api/v1/chat` | JWT | Route message through the orchestrator; returns `ChatResponse` or `ExecutionPlan` depending on execution mode |
-| `WS` | `/api/v1/chat/stream` | JWT (query param `?token=`) | Streaming chat — first frame is a `ChatRequest`, server yields text chunks, final frame is `{"done": true, "response": "...", "actions": [...]}`. 30-second heartbeat ping. |
-
-### Plans
-
-| Method | Path | Auth | Description |
-|---|---|---|---|
-| `GET` | `/api/v1/plans/playbook` | JWT | List all cached execution plan playbooks |
-| `GET` | `/api/v1/plans/playbook/{plan_id}` | JWT | Retrieve a specific playbook by ID |
-
-### Storage (Cloud Records)
-
-| Method | Path | Auth | Description |
-|---|---|---|---|
-| `POST` | `/api/v1/storage/records` | JWT | Upload an E2E encrypted record (verifies checksum, enforces storage quota) |
-| `GET` | `/api/v1/storage/records` | JWT | List record metadata with pagination (`?table`, `?page`, `?limit`); no blob bytes returned |
-| `GET` | `/api/v1/storage/records/{id}` | JWT | Download encrypted blob with `X-Checksum` response header |
-| `PUT` | `/api/v1/storage/records/{id}` | JWT | Replace an existing blob (verifies checksum, enforces quota) |
-| `DELETE` | `/api/v1/storage/records/{id}` | JWT | Delete a record and its S3 blob |
-
-### Vectors (Cloud Vector Store)
-
-| Method | Path | Auth | Description |
-|---|---|---|---|
-| `POST` | `/api/v1/storage/vectors/upsert` | JWT | Verify checksums and upsert encrypted vectors |
-| `POST` | `/api/v1/storage/vectors/search` | JWT | Search user-scoped vector namespace |
-| `DELETE` | `/api/v1/storage/vectors` | JWT | Delete vectors by ID list |
-
-### Backup
-
-| Method | Path | Auth | Description |
-|---|---|---|---|
-| `PUT` | `/api/v1/backup` | JWT | Upload encrypted backup blob with custom headers (`X-Backup-Version`, `X-Backup-Timestamp`, `X-Backup-Checksum`). Tier quota enforced. |
-| `GET` | `/api/v1/backup` | JWT | Download latest backup blob. Supports `If-Modified-Since`. |
-| `GET` | `/api/v1/backup/history` | JWT | List backup metadata (no blob content) |
-| `DELETE` | `/api/v1/backup/{backup_id}` | JWT | Delete a specific backup |
-
-### Plugins (Marketplace)
-
-| Method | Path | Auth | Description |
-|---|---|---|---|
-| `GET` | `/api/v1/plugins` | JWT (Power+) | Browse the marketplace (`?category`, `?q`, `?page`, `?sort=rating\|installs\|newest`) |
-| `GET` | `/api/v1/plugins/{id}` | JWT (Power+) | Plugin detail with install count and ratings |
-| `POST` | `/api/v1/plugins/{id}/install` | JWT (Power+) | Install plugin; triggers Stripe Connect revenue split for paid plugins |
-| `DELETE` | `/api/v1/plugins/{id}/install` | JWT | Uninstall plugin |
-
-### Billing
-
-| Method | Path | Auth | Description |
-|---|---|---|---|
-| `POST` | `/api/v1/billing/checkout` | JWT | Create a Stripe checkout session, returns `{"checkout_url": "..."}` |
-| `POST` | `/api/v1/billing/webhook` | Stripe signature | Handle Stripe events: `checkout.session.completed`, `customer.subscription.updated`, `customer.subscription.deleted`, `invoice.payment_failed` |
-| `GET` | `/api/v1/billing/subscription` | JWT | Get current subscription information |
-| `DELETE` | `/api/v1/billing/subscription` | JWT | Cancel subscription and revert to free tier |
-
----
-
-## Data Model
-
-9 tables managed by Alembic migrations. Source: `app/models.py`
-
-### Tables
-
-| Table | Primary Key | Key Columns | Purpose |
-|---|---|---|---|
-| `users` | `id` (UUID) | `email` (unique), `password_hash`, `tier`, `stripe_customer_id`, timestamps | User accounts |
-| `refresh_tokens` | `id` (UUID) | `user_id` (FK), `token_hash` (SHA-256, unique), `expires_at` | Hashed refresh tokens for rotation |
-| `subscriptions` | `id` (UUID) | `user_id` (FK, unique), `stripe_subscription_id`, `tier`, `status`, `current_period_end` | Stripe subscription records |
-| `storage_records` | `id` (UUID) | `user_id` (FK), `table_name`, `s3_key`, `checksum`, `size_bytes`, timestamps | S3 blob metadata (no plaintext content) |
-| `backup_metadata` | `id` (UUID) | `user_id` (FK), `s3_key`, `version`, `timestamp`, `checksum`, `size_bytes` | Backup manifests |
-| `plugins` | `id` (String) | `name`, `description`, `version`, `author_id` (FK), `category`, `price_cents`, `permissions` (JSON), `status`, `s3_package_key`, `install_count`, `avg_rating` | Marketplace plugin catalog |
-| `plugin_installations` | `id` (UUID) | `plugin_id` (FK), `user_id` (FK), unique constraint on (`plugin_id`, `user_id`) | Per-user install tracking |
-| `plugin_reviews` | `id` (UUID) | `plugin_id` (FK), `reviewer_id` (FK), `decision`, `notes`, `reviewed_at` | Admin review decisions |
-| `revenue_events` | `id` (UUID) | `plugin_id` (FK), `user_id` (FK), `amount_cents`, `developer_share_cents`, `stripe_transfer_id` | 70/30 revenue split ledger |
-
-### Enum Types
-
-| Enum | Values |
-|---|---|
-| `billing_tier` | `free`, `pro`, `power`, `team` |
-| `plugin_status` | `pending_review`, `approved`, `rejected` |
-| `review_decision` | `approved`, `rejected` |
-
-### Migrations
-
-| Version | Description |
-|---|---|
-| `001_initial_schema` | Creates all 9 tables with indexes and foreign key constraints |
-| `002_seed_plugins` | Seeds 3 approved plugins: GitHub Sync (free), Slack Notifier (€4.99), Time Tracker (€9.99) |
-
----
-
-## AI Agent System
-
-The agent system uses a registry pattern with LangChain tool-calling agents powered by GPT-4o. Source: `app/agents/`, `app/core/agent_registry.py`
-
-### Architecture
-
-- **`BaseAgent`** — Abstract base with `user_id`, `shared_memory`, and `vector_store_context`.
-- **`ChatAgent(BaseAgent)`** — Abstract `handle(query, context)` and `get_tools()` methods, plus a shared `_tool_loop(llm, messages, tools, max_iter=5)` for iterative tool calling.
-- **`AgentRegistry`** — Singleton registry with `@register` decorator, `get(name)`, `list_agents()`, and `call_agent(name, query, context)`.
-
-### Registered Agents
-
-| Agent | Registry Name | Tools | Description |
-|---|---|---|---|
-| **TaskAgent** | `task_agent` | 8 | Full task and comment CRUD. Status: `todo` / `in_progress` / `done`. Priority: `high` / `medium` / `low`. Tools: `list_tasks`, `create_task`, `update_task`, `delete_task`, `list_tasks_due_today`, `list_task_comments`, `add_task_comment`, `delete_task_comment` |
-| **ProjectAgent** | `project_agent` | 6 | Project lifecycle management. Status: `active` / `archived`. Prefers archiving over deletion. Tools: `list_projects`, `list_all_projects`, `get_project`, `create_project`, `update_project`, `delete_project` |
-| **TimelineAgent** | `timeline_agent` | 4 | Project milestones. Requires `project_id` for creation. Supports AI-suggestion and approval workflows. Tools: `list_timelines`, `create_timeline`, `update_timeline`, `delete_timeline` |
-| **NoteAgent** | `note_agent` | 5 | Markdown note management. Optionally linked to projects. Tools: `list_notes`, `get_note`, `create_note`, `update_note`, `delete_note` |
-
-All agents use the model configured by `LLM_MODEL` (default: GPT-4o) with `temperature=0` via LiteLLM. Tools return JSON action descriptors that the Electron client interprets and applies locally.
-
-### Switching LLM Providers
-
-The backend uses **LiteLLM** as a universal LLM gateway. All agents and the orchestrator instantiate models through a centralized factory in `app/core/llm.py`. To switch providers, change environment variables — no code changes required:
-
-```bash
-# OpenAI (default)
-LLM_MODEL=gpt-4o
-LLM_ROUTER_MODEL=gpt-4o-mini
-
-# Anthropic
-LLM_MODEL=anthropic/claude-3.5-sonnet
-LLM_ROUTER_MODEL=anthropic/claude-3-haiku
-
-# Google Gemini
-LLM_MODEL=gemini/gemini-pro
-LLM_ROUTER_MODEL=gemini/gemini-flash
-
-# Local Ollama
-LLM_MODEL=ollama/llama3
-LLM_ROUTER_MODEL=ollama/llama3
-
-# AWS Bedrock
-LLM_MODEL=bedrock/anthropic.claude-v2
-LLM_ROUTER_MODEL=bedrock/anthropic.claude-instant-v1
-```
-
-See the [LiteLLM provider docs](https://docs.litellm.ai/docs/providers) for the full list of 100+ supported providers and model naming conventions.
-
----
-
-## Orchestration & Execution Plans
-
-Source: `app/core/orchestrator.py`, `app/core/execution_plan.py`
-
-### Orchestrator
-
-1. **`classify_intent(message, context, registry)`** — Uses the router model (`LLM_ROUTER_MODEL`, default: GPT-4o-mini) to determine which agent should handle a message. Falls back to `task_agent` when classification is ambiguous.
-2. **`route_single(agent_name, message, context)`** — Routes to a single agent and returns a `ChatResponse`.
-3. **`route_pipeline(agent_names, message, context)`** — Executes agents sequentially; each receives `previous_results` from earlier agents. A final LLM synthesis step merges all results.
-4. **`orchestrate(request)`** — Main entry point. In `direct` mode, returns a `ChatResponse`. In `plan` mode, returns an `ExecutionPlan`.
-5. **`orchestrate_stream(request)`** — Streaming variant that yields 50-character text chunks with a final JSON frame.
-
-### Execution Plans
-
-- **`PromptTemplateRegistry`** — Maps template IDs to server-side prompt text. Clients only ever see opaque IDs, never raw prompts.
-- **`ExecutionPlanBuilder`** — Fluent builder API: `add_step()`, `add_llm_step(template_id, vars)`, `add_data_step(action, data_from_step)`. Validates step references on `build()`.
-- **`PlanCache`** — LRU cache (maxsize 1000) for storing plans as reusable playbooks.
-
-### Built-in Templates (6)
-
-`tpl_task_agent_default`, `tpl_timeline_agent_default`, `tpl_project_agent_default`, `tpl_note_agent_default`, `tpl_task_extract_from_project`, `tpl_note_weekly_summary`
-
-### Built-in Playbooks (2)
-
-| Playbook | Description |
-|---|---|
-| `create_tasks_from_project` | LLM extracts actionable tasks from project context, then creates task records |
-| `generate_weekly_note` | LLM generates a weekly summary, then creates a note record |
-
----
-
-## Middleware
-
-Middleware executes in this order on each request: **TierRateLimit → Sanitizer → CORS → Router**
-
-### JWT Authentication
-
-Source: `app/api/middleware/auth.py`
-
-- FastAPI dependency `get_current_user` validates the `Bearer` JWT and extracts `user_id` and `email`.
-- **Live tier lookup** — The current tier is fetched from the `subscriptions` table on every request (not cached in the JWT), so upgrades and downgrades take immediate effect.
-- Falls back to `free` when no subscription row exists.
-- Raises `401 Unauthorized` on invalid or expired tokens.
-- **Exempt paths:** `/api/v1/auth/register`, `/api/v1/auth/login`, `/api/v1/billing/webhook`
-
-### Tier-Based Rate Limiter
-
-Source: `app/api/middleware/rate_limit.py`
-
-- `TierRateLimitMiddleware` — Sliding-window in-process rate limiter (no Redis dependency).
-- Per-user 60-second window sized by subscription tier:
-
-| Tier | Requests / Minute |
-|---|---|
-| Free | 20 |
-| Pro | 60 |
-| Power | 120 |
-| Team | 200 |
-
-- Returns `429 Too Many Requests` with a `Retry-After` header when the limit is exceeded.
-- **Exempt paths:** register, login, webhook, health
-
-### Response Sanitizer
-
-Source: `app/api/middleware/sanitizer.py`
-
-- Runs only on `/api/v1/chat` endpoints.
-- Scans JSON response bodies and replaces leaked prompt IP fragments with `[REDACTED]`.
-- Detects: system prompt openers, agent routing metadata, LangChain tool schemas, internal reasoning markers (`<thinking>`, `[INST]`), and known prompt fingerprints.
-- Logs sanitization events as `WARNING`.
-- Binary responses (storage, backup) are never touched.
-
----
-
-## Storage Layer
-
-### Blob Store
-
-Source: `app/storage/blob_store.py`
-
-- S3-backed storage for E2E encrypted blobs.
-- Object keys follow the pattern: `{user_id}/{table}/{record_id}`
-- Server-side SSE-S3 encryption at rest (additional layer on top of client-side E2E encryption).
-- Methods: `upload()`, `download()`, `delete()` (idempotent), `list_keys()`
-- The backend **never inspects or decrypts blob content**.
-
-### Vector Store
-
-Source: `app/storage/vector_store.py`
-
-- Runtime-configurable: **Pinecone** (when `PINECONE_API_KEY` is set) or **Qdrant** (fallback).
-- User isolation: Pinecone uses `namespace=user_id`; Qdrant filters by `user_id` payload field.
-- 32-dimensional SHA-256-derived float vectors (deterministic, not semantically meaningful on encrypted data — a documented trade-off for privacy).
-- Encrypted blobs are stored as base64 in metadata/payload for verbatim retrieval.
-- Methods: `upsert()`, `search()`, `delete()`
-
-### Encryption Utilities
-
-Source: `app/storage/encryption.py`
-
-- `verify_checksum(blob, checksum)` — SHA-256 hash comparison using `hmac.compare_digest` (constant-time to prevent timing attacks).
-- `reject_if_tampered(blob, checksum)` — Raises HTTP 400 on checksum mismatch.
-- **No decryption key ever reaches the backend.**
-
----
-
-## Billing & Tiers
-
-Source: `app/billing/stripe_service.py`, `app/billing/tier_manager.py`
-
-### Feature Matrix
-
-| Feature | Free | Pro | Power | Team |
-|---|---|---|---|---|
-| AI Agents | 3 | Unlimited | Unlimited | Unlimited |
-| Batch Active | 2 | 10 | Unlimited | Unlimited |
-| Cloud Storage | 0 GB | 5 GB | 25 GB | Unlimited |
-| Backup Storage | 0 GB | 5 GB | 25 GB | Unlimited |
-| LLM Providers | 1 | Unlimited | Unlimited | Unlimited |
-| Batch Builder | — | — | ✓ | ✓ |
-| Plugin Marketplace | — | — | ✓ | ✓ |
-| SSO | — | — | — | ✓ |
-| Rate Limit | 20 req/min | 60 req/min | 120 req/min | 200 req/min |
-
-### Stripe Integration
-
-- **Checkout** — `create_checkout_session(user_id, tier)` creates a Stripe Checkout session. Returns a stub URL when Stripe is not configured.
-- **Webhooks** — Handles `checkout.session.completed`, `customer.subscription.updated`, `customer.subscription.deleted`, and `invoice.payment_failed`.
-- **Subscription management** — `get_subscription()` returns the current subscription record; `cancel_subscription()` cancels via the Stripe API and reverts the user to the free tier.
-- **Price IDs:** `price_pro_monthly`, `price_power_monthly`, `price_team_monthly`
-
-### Tier Manager
-
-- `get_tier(user_id)` — Returns the user's current billing tier.
-- `check_feature(tier, feature)` — Boolean feature gate check.
-- `require_feature(tier, feature)` — Raises HTTP 403 if the feature is not available.
-- `enforce_quota(user_id, tier)` / `enforce_backup_quota(user_id, tier)` — Raises HTTP 402 if storage limits are exceeded.
-
----
-
-## Plugin Marketplace
-
-Source: `app/marketplace/`
-
-### Plugin Registry
-
-- PostgreSQL-backed catalog of submitted and approved plugins.
-- `list_plugins(db, category, query, page, sort)` — Paginated listing (page size: 20) with optional filtering by category, text search, and sorting by `rating`, `installs`, or `newest`.
-- `get_plugin(db, plugin_id)` — Full manifest with install count and ratings.
-- `submit_plugin(db, manifest, s3_key)` — Submits a plugin with `pending_review` status.
-- `approve_plugin()` / `reject_plugin(reason)` — Admin workflow for plugin approval.
-- `record_install()` / `record_uninstall()` — Tracks per-user installations and updates install counts.
-
-### Review Queue
-
-- Automated security checklist before human review:
-  - Plugin ID must match `^[a-z0-9-]+$`
-  - Permissions must be from the allowed set only
-  - No binary blobs in the manifest
-- **Allowed permissions:** `read:tasks`, `write:tasks`, `read:projects`, `write:projects`, `read:notes`, `write:notes`, `read:timelines`, `write:timelines`, `read:calendar`, `write:calendar`
-- `get_pending(db)` — Lists plugins awaiting review.
-- `submit_review(db, plugin_id, reviewer_id, decision, notes)` — Records the review decision.
-
-### Revenue Sharing
-
-- **70% developer / 30% platform** split on all paid plugin sales.
-- `record_install(db, plugin_id, user_id, amount_cents)` — Records the revenue event and triggers a Stripe Connect transfer for the developer share.
-- `get_earnings(db, developer_id, period)` — Aggregated earnings report for plugin developers.
-- Gracefully stubs transfers when Stripe is not configured.
-
-### Seed Plugins
-
-| Plugin | Category | Price |
-|---|---|---|
-| GitHub Sync | Productivity | Free |
-| Slack Notifier | Communication | €4.99 |
-| Time Tracker | Productivity | €9.99 |
-
----
-
-## Testing
-
-### Running Tests
-
-```bash
-# Run all tests
-pytest
-
-# Run a specific test file
-pytest tests/test_auth.py
-
-# Run with verbose output
-pytest -v
-```
-
-### Test Infrastructure
-
-- **Database:** Async SQLite in-memory via `aiosqlite` + `StaticPool` — fast, no PostgreSQL needed.
-- **S3 mock:** `moto[s3]` with a fixture that patches `BlobStore` settings.
-- **Auth helpers:** `make_jwt(tier)` and `auth_header(tier)` generate per-tier test tokens.
-- **Seed data:** Auto-creates one `User` + `Subscription` per tier (free/pro/power/team) before each test.
-- **Plugin seeds:** Fixture adds 3 approved plugins for marketplace tests.
-- **FK enforcement:** SQLite `PRAGMA foreign_keys=ON`.
-- **No external dependencies** — all tests run fully offline.
-
-### Test Coverage
-
-| File | Coverage |
-|---|---|
-| `test_auth.py` | Register, login, token access, refresh, expiration |
-| `test_orchestrator.py` | Intent classification, single agent routing, pipeline, plan mode |
-| `test_agents.py` | Each agent with mocked LLM: registration, tools, handle method |
-| `test_storage.py` | Create, list, download, update, delete records; checksum rejection; quota enforcement |
-| `test_backup.py` | Upload, download, history, delete; tier-based storage limits |
-| `test_plugins.py` | List, install, uninstall, revenue events, tier gate enforcement |
-| `test_agent_registry.py` | Registry singleton, registration, lookup, listing |
-| `test_execution_plan.py` | Plan builder, template registry, plan cache |
-| `test_middleware.py` | Rate limiting by tier, sanitizer prompt leak detection |
-
----
-
-## Project Structure
-
-```
-adiuva-api/
-├── alembic.ini                  # Alembic configuration
-├── BACKEND_PLAN.md              # Architecture & design decisions
-├── docker-compose.yml           # Docker Compose (app + PostgreSQL)
-├── Dockerfile                   # Multi-stage production build
-├── requirements.txt             # Python dependencies
-│
-├── alembic/                     # Database migrations
-│   ├── env.py                   # Alembic environment config
-│   ├── script.py.mako           # Migration template
-│   └── versions/
-│       ├── 001_initial_schema.py    # Tables, indexes, FKs
-│       └── 002_seed_plugins.py      # Seed marketplace plugins
-│
-├── app/                         # Application source
-│   ├── main.py                  # FastAPI app factory, middleware, routes
-│   ├── db.py                    # Async SQLAlchemy engine & session
-│   ├── models.py                # SQLAlchemy ORM models (9 tables)
-│   ├── schemas.py               # Pydantic request/response schemas
-│   │
-│   ├── config/
-│   │   └── settings.py          # Pydantic Settings (env vars)
-│   │
-│   ├── agents/                  # LLM-powered domain agents
-│   │   ├── task_agent.py        # Task & comment CRUD (8 tools)
-│   │   ├── project_agent.py     # Project lifecycle (6 tools)
-│   │   ├── timeline_agent.py  # Milestones (4 tools)
-│   │   └── note_agent.py        # Markdown notes (5 tools)
-│   │
-│   ├── core/                    # Orchestration engine
-│   │   ├── agent_registry.py    # BaseAgent, ChatAgent, AgentRegistry
-│   │   ├── llm.py               # LiteLLM factory (get_llm, get_router_llm)
-│   │   ├── orchestrator.py      # Intent classification & routing
-│   │   └── execution_plan.py    # Plan builder, templates, cache
-│   │
-│   ├── api/                     # HTTP layer
-│   │   ├── deps.py              # Shared FastAPI dependencies
-│   │   ├── middleware/
-│   │   │   ├── auth.py          # JWT validation, live tier lookup
-│   │   │   ├── rate_limit.py    # Sliding-window tier rate limiter
-│   │   │   └── sanitizer.py     # Prompt IP leak protection
-│   │   └── routes/
-│   │       ├── auth.py          # Register, login, refresh, me
-│   │       ├── chat.py          # Chat + WebSocket streaming
-│   │       ├── plans.py         # Execution plan playbooks
-│   │       ├── storage.py       # E2E encrypted record CRUD
-│   │       ├── vectors.py       # Vector upsert, search, delete
-│   │       ├── backup.py        # Encrypted backup management
-│   │       ├── plugins.py       # Marketplace browse & install
-│   │       └── billing.py       # Stripe checkout & webhooks
-│   │
-│   ├── storage/                 # Storage backends
-│   │   ├── blob_store.py        # S3 blob storage
-│   │   ├── vector_store.py      # Pinecone / Qdrant vector store
-│   │   └── encryption.py        # Checksum verification utilities
-│   │
-│   ├── billing/                 # Subscription management
-│   │   ├── stripe_service.py    # Stripe API integration
-│   │   └── tier_manager.py      # Feature matrix & quota enforcement
-│   │
-│   └── marketplace/             # Plugin ecosystem
-│       ├── plugin_registry.py   # Catalog CRUD & search
-│       ├── plugin_review.py     # Security checklist & review queue
-│       └── revenue_share.py     # 70/30 split & Stripe Connect
-│
-└── tests/                       # Test suite
-    ├── conftest.py              # Fixtures: DB, S3, auth, seeds
-    ├── test_auth.py
-    ├── test_orchestrator.py
-    ├── test_agents.py
-    ├── test_storage.py
-    ├── test_backup.py
-    ├── test_plugins.py
-    ├── test_agent_registry.py
-    ├── test_execution_plan.py
-    └── test_middleware.py
-```
-
----
-
-## License
-
-*To be determined.*

alembic/env.py

@@ -16,7 +16,7 @@ import re
 from logging.config import fileConfig
 
 from alembic import context
-from sqlalchemy import engine_from_config, pool
+from sqlalchemy import pool
 from sqlalchemy.ext.asyncio import create_async_engine
 
 # Alembic Config object (gives access to alembic.ini values).

alembic/versions/001_initial_schema.py

@@ -1,5 +1,4 @@
-"""Initial schema: users, refresh_tokens, subscriptions, storage_records,
+"""Initial schema: users, refresh_tokens, subscriptions.
-backup_metadata, plugins, plugin_installations, plugin_reviews, revenue_events.
 
 Revision ID: 001
 Revises:
@@ -28,18 +27,6 @@ def upgrade() -> None:
         EXCEPTION WHEN duplicate_object THEN NULL;
         END $$;
     """)
-    op.execute("""
-        DO $$ BEGIN
-            CREATE TYPE plugin_status AS ENUM ('pending_review', 'approved', 'rejected');
-        EXCEPTION WHEN duplicate_object THEN NULL;
-        END $$;
-    """)
-    op.execute("""
-        DO $$ BEGIN
-            CREATE TYPE review_decision AS ENUM ('approved', 'rejected');
-        EXCEPTION WHEN duplicate_object THEN NULL;
-        END $$;
-    """)
 
     # ── users ─────────────────────────────────────────────────────────────
     op.create_table(
@@ -88,122 +75,10 @@ def upgrade() -> None:
     op.create_index("ix_subscriptions_user_id", "subscriptions", ["user_id"])
     op.create_index("ix_subscriptions_stripe_id", "subscriptions", ["stripe_subscription_id"])
 
-    # ── storage_records ───────────────────────────────────────────────────
-    op.create_table(
-        "storage_records",
-        sa.Column("id", postgresql.UUID(as_uuid=False), nullable=False),
-        sa.Column("user_id", postgresql.UUID(as_uuid=False), nullable=False),
-        sa.Column("table_name", sa.String(100), nullable=False),
-        sa.Column("s3_key", sa.String(500), nullable=False),
-        sa.Column("checksum", sa.String(64), nullable=False),
-        sa.Column("size_bytes", sa.Integer, nullable=False),
-        sa.Column("created_at", sa.DateTime(timezone=True), nullable=False, server_default=sa.text("now()")),
-        sa.Column("updated_at", sa.DateTime(timezone=True), nullable=False, server_default=sa.text("now()")),
-        sa.PrimaryKeyConstraint("id"),
-        sa.ForeignKeyConstraint(["user_id"], ["users.id"], ondelete="CASCADE"),
-    )
-    op.create_index("ix_storage_records_user_id", "storage_records", ["user_id"])
-
-    # ── backup_metadata ───────────────────────────────────────────────────
-    op.create_table(
-        "backup_metadata",
-        sa.Column("id", postgresql.UUID(as_uuid=False), nullable=False),
-        sa.Column("user_id", postgresql.UUID(as_uuid=False), nullable=False),
-        sa.Column("s3_key", sa.String(500), nullable=False),
-        sa.Column("version", sa.Integer, nullable=False),
-        sa.Column("timestamp", sa.BigInteger, nullable=False),
-        sa.Column("checksum", sa.String(64), nullable=False),
-        sa.Column("size_bytes", sa.Integer, nullable=False),
-        sa.Column("created_at", sa.DateTime(timezone=True), nullable=False, server_default=sa.text("now()")),
-        sa.PrimaryKeyConstraint("id"),
-        sa.ForeignKeyConstraint(["user_id"], ["users.id"], ondelete="CASCADE"),
-    )
-    op.create_index("ix_backup_metadata_user_id", "backup_metadata", ["user_id"])
-
-    # ── plugins ───────────────────────────────────────────────────────────
-    op.create_table(
-        "plugins",
-        sa.Column("id", sa.String(255), nullable=False),
-        sa.Column("name", sa.String(255), nullable=False),
-        sa.Column("description", sa.Text, nullable=False, server_default=""),
-        sa.Column("version", sa.String(50), nullable=False, server_default="1.0.0"),
-        sa.Column("author_id", postgresql.UUID(as_uuid=False), nullable=True),
-        sa.Column("author_name", sa.String(255), nullable=False, server_default=""),
-        sa.Column("category", sa.String(100), nullable=False, server_default=""),
-        sa.Column("price_cents", sa.Integer, nullable=False, server_default="0"),
-        sa.Column("permissions", sa.Text, nullable=False, server_default="[]"),
-        sa.Column("status", postgresql.ENUM("pending_review", "approved", "rejected", name="plugin_status", create_type=False), nullable=False, server_default="pending_review"),
-        sa.Column("s3_package_key", sa.String(500), nullable=True),
-        sa.Column("install_count", sa.Integer, nullable=False, server_default="0"),
-        sa.Column("avg_rating", sa.Float, nullable=False, server_default="0.0"),
-        sa.Column("rejection_reason", sa.Text, nullable=True),
-        sa.Column("submitted_at", sa.DateTime(timezone=True), nullable=False, server_default=sa.text("now()")),
-        sa.Column("created_at", sa.DateTime(timezone=True), nullable=False, server_default=sa.text("now()")),
-        sa.PrimaryKeyConstraint("id"),
-        sa.ForeignKeyConstraint(["author_id"], ["users.id"], ondelete="SET NULL"),
-    )
-
-    # ── plugin_installations ──────────────────────────────────────────────
-    op.create_table(
-        "plugin_installations",
-        sa.Column("id", postgresql.UUID(as_uuid=False), nullable=False),
-        sa.Column("plugin_id", sa.String(255), nullable=False),
-        sa.Column("user_id", postgresql.UUID(as_uuid=False), nullable=False),
-        sa.Column("installed_at", sa.DateTime(timezone=True), nullable=False, server_default=sa.text("now()")),
-        sa.PrimaryKeyConstraint("id"),
-        sa.ForeignKeyConstraint(["plugin_id"], ["plugins.id"], ondelete="CASCADE"),
-        sa.ForeignKeyConstraint(["user_id"], ["users.id"], ondelete="CASCADE"),
-        sa.UniqueConstraint("plugin_id", "user_id", name="uq_plugin_user"),
-    )
-    op.create_index("ix_plugin_installations_plugin_id", "plugin_installations", ["plugin_id"])
-    op.create_index("ix_plugin_installations_user_id", "plugin_installations", ["user_id"])
-
-    # ── plugin_reviews ────────────────────────────────────────────────────
-    op.create_table(
-        "plugin_reviews",
-        sa.Column("id", postgresql.UUID(as_uuid=False), nullable=False),
-        sa.Column("plugin_id", sa.String(255), nullable=False),
-        sa.Column("reviewer_id", postgresql.UUID(as_uuid=False), nullable=True),
-        sa.Column("decision", postgresql.ENUM("approved", "rejected", name="review_decision", create_type=False), nullable=False),
-        sa.Column("notes", sa.Text, nullable=True),
-        sa.Column("reviewed_at", sa.DateTime(timezone=True), nullable=False, server_default=sa.text("now()")),
-        sa.Column("created_at", sa.DateTime(timezone=True), nullable=False, server_default=sa.text("now()")),
-        sa.PrimaryKeyConstraint("id"),
-        sa.ForeignKeyConstraint(["plugin_id"], ["plugins.id"], ondelete="CASCADE"),
-        sa.ForeignKeyConstraint(["reviewer_id"], ["users.id"], ondelete="SET NULL"),
-    )
-    op.create_index("ix_plugin_reviews_plugin_id", "plugin_reviews", ["plugin_id"])
-
-    # ── revenue_events ────────────────────────────────────────────────────
-    op.create_table(
-        "revenue_events",
-        sa.Column("id", postgresql.UUID(as_uuid=False), nullable=False),
-        sa.Column("plugin_id", sa.String(255), nullable=False),
-        sa.Column("user_id", postgresql.UUID(as_uuid=False), nullable=False),
-        sa.Column("amount_cents", sa.Integer, nullable=False, server_default="0"),
-        sa.Column("developer_share_cents", sa.Integer, nullable=False, server_default="0"),
-        sa.Column("stripe_transfer_id", sa.String(255), nullable=True),
-        sa.Column("paid_at", sa.DateTime(timezone=True), nullable=True),
-        sa.Column("created_at", sa.DateTime(timezone=True), nullable=False, server_default=sa.text("now()")),
-        sa.PrimaryKeyConstraint("id"),
-        sa.ForeignKeyConstraint(["plugin_id"], ["plugins.id"], ondelete="CASCADE"),
-        sa.ForeignKeyConstraint(["user_id"], ["users.id"], ondelete="CASCADE"),
-    )
-    op.create_index("ix_revenue_events_plugin_id", "revenue_events", ["plugin_id"])
-    op.create_index("ix_revenue_events_user_id", "revenue_events", ["user_id"])
-
 
 def downgrade() -> None:
-    op.drop_table("revenue_events")
-    op.drop_table("plugin_reviews")
-    op.drop_table("plugin_installations")
-    op.drop_table("plugins")
-    op.drop_table("backup_metadata")
-    op.drop_table("storage_records")
     op.drop_table("subscriptions")
     op.drop_table("refresh_tokens")
     op.drop_table("users")
 
-    op.execute("DROP TYPE IF EXISTS review_decision")
-    op.execute("DROP TYPE IF EXISTS plugin_status")
     op.execute("DROP TYPE IF EXISTS billing_tier")

alembic/versions/002_seed_plugins.py

@@ -1,92 +0,0 @@
-"""Seed approved plugins: GitHub Sync, Slack Notifier, Time Tracker.
-
-Revision ID: 002
-Revises: 001
-Create Date: 2026-03-03
-"""
-
-from __future__ import annotations
-
-import json
-from datetime import datetime, timezone
-from typing import Sequence, Union
-
-import sqlalchemy as sa
-from alembic import op
-
-revision: str = "002"
-down_revision: Union[str, None] = "001"
-branch_labels: Union[str, Sequence[str], None] = None
-depends_on: Union[str, Sequence[str], None] = None
-
-_SEED_PLUGINS = [
-    {
-        "id": "plugin-github-sync",
-        "name": "GitHub Sync",
-        "description": "Sync tasks with GitHub Issues and pull requests.",
-        "version": "1.0.0",
-        "author_name": "Adiuva",
-        "category": "productivity",
-        "price_cents": 0,
-        "permissions": json.dumps(["read:tasks", "write:tasks"]),
-        "status": "approved",
-        "s3_package_key": "plugins/plugin-github-sync/1.0.0/package.zip",
-        "install_count": 0,
-        "avg_rating": 0.0,
-    },
-    {
-        "id": "plugin-slack-notify",
-        "name": "Slack Notifier",
-        "description": "Post task and timeline updates to Slack channels.",
-        "version": "1.2.0",
-        "author_name": "Adiuva",
-        "category": "communication",
-        "price_cents": 499,
-        "permissions": json.dumps(["read:tasks", "read:timelines"]),
-        "status": "approved",
-        "s3_package_key": "plugins/plugin-slack-notify/1.2.0/package.zip",
-        "install_count": 0,
-        "avg_rating": 0.0,
-    },
-    {
-        "id": "plugin-time-tracker",
-        "name": "Time Tracker",
-        "description": "Track time spent on tasks with automatic reporting.",
-        "version": "0.9.1",
-        "author_name": "Third Party",
-        "category": "productivity",
-        "price_cents": 999,
-        "permissions": json.dumps(["read:tasks", "write:tasks"]),
-        "status": "approved",
-        "s3_package_key": "plugins/plugin-time-tracker/0.9.1/package.zip",
-        "install_count": 0,
-        "avg_rating": 0.0,
-    },
-]
-
-
-def upgrade() -> None:
-    plugins = sa.table(
-        "plugins",
-        sa.column("id", sa.String),
-        sa.column("name", sa.String),
-        sa.column("description", sa.Text),
-        sa.column("version", sa.String),
-        sa.column("author_name", sa.String),
-        sa.column("category", sa.String),
-        sa.column("price_cents", sa.Integer),
-        sa.column("permissions", sa.Text),
-        sa.column("status", sa.Enum("pending_review", "approved", "rejected", name="plugin_status")),
-        sa.column("s3_package_key", sa.String),
-        sa.column("install_count", sa.Integer),
-        sa.column("avg_rating", sa.Float),
-    )
-    op.bulk_insert(plugins, _SEED_PLUGINS)
-
-
-def downgrade() -> None:
-    op.execute(
-        "DELETE FROM plugins WHERE id IN ("
-        "'plugin-github-sync', 'plugin-slack-notify', 'plugin-time-tracker'"
-        ")"
-    )

alembic/versions/003_agent_tables.py

@@ -14,7 +14,7 @@ from alembic import op
 from sqlalchemy.dialects import postgresql
 
 revision: str = "003"
-down_revision: Union[str, None] = "002"
+down_revision: Union[str, None] = "001"
 branch_labels: Union[str, Sequence[str], None] = None
 depends_on: Union[str, Sequence[str], None] = None
 

alembic/versions/005_associative_pgvector.py

@@ -0,0 +1,54 @@
+"""Phase 1 — confirm pgvector activation on memory_associative.
+
+Migration 004 created the embedding column as vector(1536) and added the
+IVFFlat index.  This migration is the Phase-1 checkpoint:
+  1. Ensures the pgvector extension is enabled (idempotent).
+  2. Ensures the canonical Phase-1 IVFFlat index exists under the name
+     memory_associative_embedding_idx (creates it only if absent).
+
+Revision ID: 005
+Revises: 9a1f2d0b6c7e
+Create Date: 2026-04-15
+"""
+
+from __future__ import annotations
+
+from typing import Sequence, Union
+
+from alembic import op
+
+revision: str = "005"
+down_revision: Union[str, None] = "e04100e88ace"
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def upgrade() -> None:
+    # Ensure pgvector extension is enabled (also done in 004, idempotent).
+    op.execute("CREATE EXTENSION IF NOT EXISTS vector;")
+
+    # Ensure the canonical Phase-1 IVFFlat index exists.
+    # 004 may have created ix_memory_associative_embedding; this adds the
+    # Phase-1 name memory_associative_embedding_idx if it is missing.
+    op.execute(
+        """
+        DO $$
+        BEGIN
+            IF NOT EXISTS (
+                SELECT 1
+                FROM   pg_indexes
+                WHERE  tablename  = 'memory_associative'
+                  AND  indexname  = 'memory_associative_embedding_idx'
+            ) THEN
+                CREATE INDEX memory_associative_embedding_idx
+                ON memory_associative
+                USING ivfflat (embedding vector_cosine_ops)
+                WITH  (lists = 100);
+            END IF;
+        END $$;
+        """
+    )
+
+
+def downgrade() -> None:
+    op.execute("DROP INDEX IF EXISTS memory_associative_embedding_idx;")

alembic/versions/006_memory_relations.py

@@ -0,0 +1,74 @@
+"""Add memory_relations table (Phase 3 — relational tier).
+
+Revision ID: 006
+Revises: 1f5975a4f3f4
+Create Date: 2026-04-16
+"""
+
+from __future__ import annotations
+
+from typing import Sequence, Union
+
+import sqlalchemy as sa
+from alembic import op
+from sqlalchemy.dialects import postgresql
+
+revision: str = "006"
+down_revision: Union[str, None] = "1f5975a4f3f4"
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def upgrade() -> None:
+    op.create_table(
+        "memory_relations",
+        sa.Column("id", postgresql.UUID(as_uuid=False), primary_key=True),
+        sa.Column(
+            "user_id",
+            postgresql.UUID(as_uuid=False),
+            sa.ForeignKey("users.id", ondelete="CASCADE"),
+            nullable=False,
+        ),
+        sa.Column("subject_label", sa.String(128), nullable=False),
+        sa.Column("subject_type", sa.String(32), nullable=False),
+        sa.Column("predicate", sa.String(64), nullable=False),
+        sa.Column("object_label", sa.String(128), nullable=False),
+        sa.Column("object_type", sa.String(32), nullable=False),
+        sa.Column("confidence", sa.Float, nullable=False, server_default="0.7"),
+        sa.Column(
+            "source_episode_id",
+            postgresql.UUID(as_uuid=False),
+            sa.ForeignKey("memory_episodic.id", ondelete="SET NULL"),
+            nullable=True,
+        ),
+        sa.Column("notes_encrypted", sa.LargeBinary, nullable=True),
+        sa.Column(
+            "created_at",
+            sa.DateTime(timezone=True),
+            nullable=False,
+            server_default=sa.func.now(),
+        ),
+        sa.Column(
+            "updated_at",
+            sa.DateTime(timezone=True),
+            nullable=False,
+            server_default=sa.func.now(),
+        ),
+        sa.Column("last_confirmed_at", sa.DateTime(timezone=True), nullable=True),
+    )
+    op.create_index(
+        "memory_relations_user_subject_idx",
+        "memory_relations",
+        ["user_id", "subject_label"],
+    )
+    op.create_index(
+        "memory_relations_user_predicate_idx",
+        "memory_relations",
+        ["user_id", "predicate"],
+    )
+
+
+def downgrade() -> None:
+    op.drop_index("memory_relations_user_predicate_idx", "memory_relations")
+    op.drop_index("memory_relations_user_subject_idx", "memory_relations")
+    op.drop_table("memory_relations")

alembic/versions/1f5975a4f3f4_add_extraction_queue.py

@@ -0,0 +1,38 @@
+"""add extraction_queue
+
+Revision ID: 1f5975a4f3f4
+Revises: 005
+Create Date: 2026-04-16 17:26:25.790870
+
+"""
+from __future__ import annotations
+
+from typing import Sequence, Union
+
+from alembic import op
+import sqlalchemy as sa
+
+
+# revision identifiers, used by Alembic.
+revision: str = '1f5975a4f3f4'
+down_revision: Union[str, None] = '005'
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def upgrade() -> None:
+    op.create_table(
+        'extraction_queue',
+        sa.Column('id', sa.Uuid(as_uuid=False), nullable=False),
+        sa.Column('user_id', sa.Uuid(as_uuid=False), nullable=False),
+        sa.Column('episode_id', sa.Uuid(as_uuid=False), nullable=True),
+        sa.Column('created_at', sa.DateTime(timezone=True), server_default=sa.text('now()'), nullable=False),
+        sa.ForeignKeyConstraint(['user_id'], ['users.id'], ondelete='CASCADE'),
+        sa.PrimaryKeyConstraint('id'),
+    )
+    op.create_index(op.f('ix_extraction_queue_user_id'), 'extraction_queue', ['user_id'], unique=False)
+
+
+def downgrade() -> None:
+    op.drop_index(op.f('ix_extraction_queue_user_id'), table_name='extraction_queue')
+    op.drop_table('extraction_queue')

alembic/versions/a3b9c0d1e2f3_add_agent_config_to_local_agents.py

@@ -0,0 +1,107 @@
+"""Restore agent config tables and add agent_config column.
+
+9a1f2d0b6c7e dropped local_agent_configs and cloud_agent_configs, but both
+ORM models are still active. This migration recreates them with agent_config
+added to local_agent_configs.
+
+Revision ID: a3b9c0d1e2f3
+Revises: 9a1f2d0b6c7e
+Create Date: 2026-04-07 00:00:00.000000
+
+"""
+from __future__ import annotations
+
+from typing import Sequence, Union
+
+import sqlalchemy as sa
+from alembic import op
+from sqlalchemy.dialects import postgresql
+
+
+# revision identifiers, used by Alembic.
+revision: str = "a3b9c0d1e2f3"
+down_revision: Union[str, None] = "9a1f2d0b6c7e"
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def upgrade() -> None:
+    # Recreate enum types (idempotent — they may already exist from migration 003)
+    op.execute("""
+        DO $$ BEGIN
+            CREATE TYPE agent_type AS ENUM ('local', 'cloud');
+        EXCEPTION WHEN duplicate_object THEN NULL;
+        END $$;
+    """)
+    op.execute("""
+        DO $$ BEGIN
+            CREATE TYPE agent_run_status AS ENUM ('running', 'success', 'error', 'partial');
+        EXCEPTION WHEN duplicate_object THEN NULL;
+        END $$;
+    """)
+    op.execute("""
+        DO $$ BEGIN
+            CREATE TYPE cloud_provider AS ENUM ('gmail', 'teams', 'outlook');
+        EXCEPTION WHEN duplicate_object THEN NULL;
+        END $$;
+    """)
+
+    bind = op.get_bind()
+    inspector = sa.inspect(bind)
+    existing = set(inspector.get_table_names())
+
+    # ── local_agent_configs (with agent_config column) ────────────────────
+    if "local_agent_configs" not in existing:
+        op.create_table(
+            "local_agent_configs",
+            sa.Column("id", postgresql.UUID(as_uuid=False), nullable=False),
+            sa.Column("user_id", postgresql.UUID(as_uuid=False), nullable=False),
+            sa.Column("device_id", sa.String(255), nullable=False),
+            sa.Column("name", sa.String(255), nullable=False),
+            sa.Column("directory_paths", sa.JSON, nullable=False, server_default="[]"),
+            sa.Column("data_types", sa.JSON, nullable=False, server_default="[]"),
+            sa.Column("prompt_template", sa.Text, nullable=False, server_default=""),
+            sa.Column("agent_config", sa.JSON, nullable=True),
+            sa.Column("file_extensions", sa.JSON, nullable=False, server_default="[]"),
+            sa.Column("schedule_cron", sa.String(100), nullable=False, server_default="0 */6 * * *"),
+            sa.Column("enabled", sa.Boolean, nullable=False, server_default=sa.true()),
+            sa.Column("last_run_at", sa.DateTime(timezone=True), nullable=True),
+            sa.Column("created_at", sa.DateTime(timezone=True), nullable=False, server_default=sa.text("now()")),
+            sa.Column("updated_at", sa.DateTime(timezone=True), nullable=False, server_default=sa.text("now()")),
+            sa.PrimaryKeyConstraint("id"),
+            sa.ForeignKeyConstraint(["user_id"], ["users.id"], ondelete="CASCADE"),
+        )
+        op.create_index("ix_local_agent_configs_user_id", "local_agent_configs", ["user_id"])
+
+    # ── cloud_agent_configs ───────────────────────────────────────────────
+    if "cloud_agent_configs" not in existing:
+        op.create_table(
+            "cloud_agent_configs",
+            sa.Column("id", postgresql.UUID(as_uuid=False), nullable=False),
+            sa.Column("user_id", postgresql.UUID(as_uuid=False), nullable=False),
+            sa.Column(
+                "provider",
+                postgresql.ENUM("gmail", "teams", "outlook", name="cloud_provider", create_type=False),
+                nullable=False,
+            ),
+            sa.Column("name", sa.String(255), nullable=False),
+            sa.Column("data_types", sa.JSON, nullable=False, server_default="[]"),
+            sa.Column("prompt_template", sa.Text, nullable=False, server_default=""),
+            sa.Column("oauth_token_encrypted", sa.Text, nullable=True),
+            sa.Column("filter_config", sa.JSON, nullable=True),
+            sa.Column("schedule_cron", sa.String(100), nullable=False, server_default="0 */6 * * *"),
+            sa.Column("enabled", sa.Boolean, nullable=False, server_default=sa.true()),
+            sa.Column("last_run_at", sa.DateTime(timezone=True), nullable=True),
+            sa.Column("created_at", sa.DateTime(timezone=True), nullable=False, server_default=sa.text("now()")),
+            sa.Column("updated_at", sa.DateTime(timezone=True), nullable=False, server_default=sa.text("now()")),
+            sa.PrimaryKeyConstraint("id"),
+            sa.ForeignKeyConstraint(["user_id"], ["users.id"], ondelete="CASCADE"),
+        )
+        op.create_index("ix_cloud_agent_configs_user_id", "cloud_agent_configs", ["user_id"])
+
+
+def downgrade() -> None:
+    op.drop_index("ix_cloud_agent_configs_user_id", table_name="cloud_agent_configs")
+    op.drop_table("cloud_agent_configs")
+    op.drop_index("ix_local_agent_configs_user_id", table_name="local_agent_configs")
+    op.drop_table("local_agent_configs")

alembic/versions/b4c0d1e2f3a4_add_oauth_and_avatar.py

@@ -0,0 +1,56 @@
+"""Add oauth_accounts table, nullable password_hash, avatar_url to users.
+
+Revision ID: b4c0d1e2f3a4
+Revises: a3b9c0d1e2f3
+Create Date: 2026-04-10 00:00:00.000000
+
+"""
+from __future__ import annotations
+
+from typing import Sequence, Union
+
+import sqlalchemy as sa
+from alembic import op
+from sqlalchemy.dialects import postgresql
+
+
+# revision identifiers, used by Alembic.
+revision: str = "b4c0d1e2f3a4"
+down_revision: Union[str, None] = "a3b9c0d1e2f3"
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def upgrade() -> None:
+    # ── users: make password_hash nullable (social users have no password) ──
+    op.alter_column("users", "password_hash", existing_type=sa.String(255), nullable=True)
+
+    # ── users: add avatar_url ─────────────────────────────────────────────
+    op.add_column("users", sa.Column("avatar_url", sa.String(2048), nullable=True))
+
+    # ── oauth_accounts ────────────────────────────────────────────────────
+    op.create_table(
+        "oauth_accounts",
+        sa.Column("id", postgresql.UUID(as_uuid=False), nullable=False),
+        sa.Column("user_id", postgresql.UUID(as_uuid=False), nullable=False),
+        sa.Column("provider", sa.String(50), nullable=False),
+        sa.Column("provider_user_id", sa.String(255), nullable=False),
+        sa.Column("provider_email", sa.String(255), nullable=True),
+        sa.Column(
+            "created_at",
+            sa.DateTime(timezone=True),
+            nullable=False,
+            server_default=sa.text("now()"),
+        ),
+        sa.PrimaryKeyConstraint("id"),
+        sa.ForeignKeyConstraint(["user_id"], ["users.id"], ondelete="CASCADE"),
+        sa.UniqueConstraint("provider", "provider_user_id", name="uq_oauth_provider_user"),
+    )
+    op.create_index("ix_oauth_accounts_user_id", "oauth_accounts", ["user_id"])
+
+
+def downgrade() -> None:
+    op.drop_index("ix_oauth_accounts_user_id", table_name="oauth_accounts")
+    op.drop_table("oauth_accounts")
+    op.drop_column("users", "avatar_url")
+    op.alter_column("users", "password_hash", existing_type=sa.String(255), nullable=False)

alembic/versions/c5d1e2f3a4b5_add_onboarding_completed_at.py

@@ -0,0 +1,31 @@
+"""Add onboarding_completed_at column to users table.
+
+Revision ID: c5d1e2f3a4b5
+Revises: b4c0d1e2f3a4
+Create Date: 2026-04-11 00:00:00.000000
+
+"""
+from __future__ import annotations
+
+from typing import Sequence, Union
+
+import sqlalchemy as sa
+from alembic import op
+
+
+# revision identifiers, used by Alembic.
+revision: str = "c5d1e2f3a4b5"
+down_revision: Union[str, None] = "b4c0d1e2f3a4"
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def upgrade() -> None:
+    op.add_column(
+        "users",
+        sa.Column("onboarding_completed_at", sa.DateTime(timezone=True), nullable=True),
+    )
+
+
+def downgrade() -> None:
+    op.drop_column("users", "onboarding_completed_at")

alembic/versions/e04100e88ace_avatar_url_varchar_to_text.py

@@ -0,0 +1,34 @@
+"""avatar_url_varchar_to_text
+
+Revision ID: e04100e88ace
+Revises: c5d1e2f3a4b5
+Create Date: 2026-04-13 09:13:06.733674
+
+"""
+from __future__ import annotations
+
+from typing import Sequence, Union
+
+from alembic import op
+import sqlalchemy as sa
+
+
+# revision identifiers, used by Alembic.
+revision: str = 'e04100e88ace'
+down_revision: Union[str, None] = 'c5d1e2f3a4b5'
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def upgrade() -> None:
+    op.alter_column('users', 'avatar_url',
+               existing_type=sa.VARCHAR(length=2048),
+               type_=sa.Text(),
+               existing_nullable=True)
+
+
+def downgrade() -> None:
+    op.alter_column('users', 'avatar_url',
+               existing_type=sa.Text(),
+               type_=sa.VARCHAR(length=2048),
+               existing_nullable=True)

app/agents/filesystem_agent.py

@@ -7,12 +7,31 @@ handles actual disk I/O and responds with ``tool_result`` frames.
 
 from __future__ import annotations
 
+import os
+import re
+from pathlib import Path
 from typing import Any
 
 from langchain_core.tools import tool
 
 from app.core.ws_context import execute_on_client
 
+# Max characters returned by read_file_content in journey (exploration) tools.
+# The journey only needs to understand file structure, not full content.
+_JOURNEY_READ_MAX_CHARS: int = 4000
+
+
+def _resolve_path(path: str, base: str) -> str:
+    """Resolve *path* against *base* when *path* is relative.
+
+    The LLM often passes ``"."`` meaning "the configured directory".
+    Without this, Electron resolves ``"."`` relative to its own CWD instead
+    of the user's chosen directory.
+    """
+    if os.path.isabs(path):
+        return path
+    return str(Path(base) / path)
+
 
 @tool
 async def list_directory(path: str) -> str:
@@ -83,3 +102,93 @@ FILESYSTEM_TOOLS: list[Any] = [
     read_file_content,
     get_file_metadata,
 ]
+
+
+def make_directory_tools(base_directory: str) -> list[Any]:
+    """Return filesystem tools that resolve relative paths against *base_directory*.
+
+    Use this instead of ``FILESYSTEM_TOOLS`` whenever you know the user's target
+    directory upfront (e.g., journey setup sessions).  Relative paths like ``"."``
+    from the LLM are resolved to the correct absolute path before being sent to
+    the Electron client, preventing it from falling back to its own CWD.
+    """
+
+    def _compact_for_journey(raw: str) -> str:
+        """Strip HTML noise and truncate for journey exploration.
+
+        The journey LLM only needs to understand file structure (headers,
+        first paragraphs).  Full CSS/style blocks are pure noise that eat
+        up context window budget.
+        """
+        text = re.sub(r"<style[^>]*>.*?</style>", "", raw, flags=re.DOTALL | re.IGNORECASE)
+        text = re.sub(r"<script[^>]*>.*?</script>", "", text, flags=re.DOTALL | re.IGNORECASE)
+        text = re.sub(r"<!--.*?-->", "", text, flags=re.DOTALL)
+        if len(text) > _JOURNEY_READ_MAX_CHARS:
+            text = text[:_JOURNEY_READ_MAX_CHARS] + "\n[…truncated for exploration]"
+        return text
+
+    @tool
+    async def list_directory(path: str) -> str:  # noqa: F811
+        """List files and folders in a local directory on the user's device.
+
+        Returns a formatted listing of entries with name, type (file/directory),
+        and full path.
+        """
+        resolved = _resolve_path(path, base_directory)
+        result = await execute_on_client(
+            action="list_directory",
+            data={"path": resolved},
+        )
+        entries: list[dict[str, Any]] = result.get("entries", [])
+        if not entries:
+            return f"Directory '{resolved}' is empty or does not exist."
+        lines: list[str] = []
+        for entry in entries:
+            entry_type = entry.get("type", "unknown")
+            entry_name = entry.get("name", "")
+            entry_path = entry.get("path", "")
+            lines.append(f"- [{entry_type}] {entry_name}  ({entry_path})")
+        return f"Directory listing for '{resolved}' ({len(entries)} entries):\n" + "\n".join(lines)
+
+    @tool
+    async def read_file_content(path: str) -> str:  # noqa: F811
+        """Read the text content of a local file on the user's device.
+
+        Returns the file content as a string.  Large files may be truncated
+        by the Electron client.
+        """
+        resolved = _resolve_path(path, base_directory)
+        result = await execute_on_client(
+            action="read_file_content",
+            data={"path": resolved},
+        )
+        content: str = result.get("content", "")
+        if not content:
+            return f"File '{resolved}' is empty or could not be read."
+        return _compact_for_journey(content)
+
+    @tool
+    async def get_file_metadata(path: str) -> str:  # noqa: F811
+        """Get metadata for a local file: size, creation date, modification date, extension.
+
+        Returns a formatted summary of the file's metadata.
+        """
+        resolved = _resolve_path(path, base_directory)
+        result = await execute_on_client(
+            action="get_file_metadata",
+            data={"path": resolved},
+        )
+        size = result.get("size", "unknown")
+        created = result.get("createdAt", "unknown")
+        modified = result.get("modifiedAt", "unknown")
+        extension = result.get("extension", "unknown")
+        name = result.get("name", resolved)
+        return (
+            f"File: {name}\n"
+            f"  Extension: {extension}\n"
+            f"  Size: {size} bytes\n"
+            f"  Created: {created}\n"
+            f"  Modified: {modified}"
+        )
+
+    return [list_directory, read_file_content, get_file_metadata]

app/agents/note_agent.py

@@ -18,21 +18,6 @@ _UUID_RE = re.compile(
 def _is_uuid(value: str) -> bool:
     return bool(_UUID_RE.match(value))
 
-NOTE_SYSTEM_PROMPT = (
-    "You are a note-taking assistant. You help users create, retrieve, update,\n"
-    "and delete Markdown notes in their workspace.\n\n"
-    "Rules:\n"
-    "  - content is always Markdown; preserve formatting when updating\n"
-    "  - project_id is optional; link a note to a project when mentioned\n"
-    "  - When updating, call get_note first if you need to read existing content\n"
-    "    before appending or replacing sections\n"
-    "  - list_notes without project_id returns all notes; scope with project_id\n"
-    "    when the user is working within a specific project\n"
-    "  - project_id must be a UUID; if you only know a project name, do not pass it as project_id\n"
-    "  - Do not fabricate note content — reflect what the user provides or what\n"
-    "    is already in the note (retrieved via get_note)."
-)
-
 
 @tool
 async def list_notes(project_id: str = "") -> str:

app/agents/project_agent.py

@@ -8,22 +8,6 @@ from langchain_core.tools import tool
 
 from app.core.ws_context import execute_on_client
 
-PROJECT_SYSTEM_PROMPT = (
-    "You are a project management assistant. You help users create, find,\n"
-    "update, and archive projects in their workspace.\n\n"
-    "Rules:\n"
-    "  - status must be one of: active, archived\n"
-    "  - client_id is optional; link to a client only when explicitly mentioned\n"
-    "  - ai_summary is populated only when the user asks for a project summary;\n"
-    "    derive it from context data — do not fabricate content\n"
-    "  - Use list_projects for scoped queries; list_all_projects only when the\n"
-    "    user wants a complete cross-client view including archived projects\n"
-    "  - get_project requires a project UUID; resolve the ID first by calling\n"
-    "    list_projects if you only have a project name\n"
-    "  - Prefer archiving (update_project status=archived) over deletion;\n"
-    "    only call delete_project when the user explicitly confirms deletion."
-)
-
 
 @tool
 async def list_projects(

app/agents/task_agent.py

@@ -18,23 +18,6 @@ _UUID_RE = re.compile(
 def _is_uuid(value: str) -> bool:
     return bool(_UUID_RE.match(value))
 
-TASK_SYSTEM_PROMPT = (
-    "You are a task management assistant for a project workspace.\n"
-    "You create, update, list, and track tasks and their comments.\n\n"
-    "Rules:\n"
-    "  - status must be one of: todo, in_progress, done\n"
-    "  - priority must be one of: high, medium, low\n"
-    "  - due_date is a Unix timestamp in milliseconds; convert human dates\n"
-    "  - assignees is a JSON-encoded array of strings (e.g. '[\"Alice\",\"Bob\"]')\n"
-    "  - project_id is optional; link to a project when the user mentions one\n"
-    "  - is_ai_suggested: 1 only when proactively proposing a task the user\n"
-    "    did not explicitly request; 0 otherwise\n"
-    "  - is_ai_suggested: 1 only when proactively proposing a task the user did not explicitly request; 0 otherwise\n"
-    "  - Use list_tasks_due_today for 'what's due today' queries\n"
-    "  - For update_task, use -1 for integer fields you do not want to change\n"
-    "  - Always confirm the action in plain, user-friendly language."
-)
-
 
 # ── Task tools ────────────────────────────────────────────────────────
 

app/agents/timeline_agent.py

@@ -17,20 +17,6 @@ _UUID_RE = re.compile(
 def _is_uuid(value: str) -> bool:
     return bool(_UUID_RE.match(value))
 
-TIMELINE_SYSTEM_PROMPT = (
-    "You are a project timeline assistant. Timelines are milestone dates that\n"
-    "track progress on a project — they are not calendar events.\n\n"
-    "Rules:\n"
-    "  - project_id is REQUIRED for every create; confirm with the user if unknown\n"
-    "  - For listing, project_id must be a UUID; never pass plain names as project_id\n"
-    "  - date is a Unix timestamp in milliseconds; convert human-readable dates\n"
-    "  - is_ai_suggested: 1 when proactively proposing a timeline, 0 otherwise\n"
-    "  - is_ai_suggested: 1 when proactively proposing a timeline, 0 otherwise\n"
-    "  - For update_timeline, use -1 for integer fields you do not want to change\n"
-    "  - Listing without a project_id returns all timelines across projects\n"
-    "  - Always echo the title and formatted date in your confirmation."
-)
-
 
 @tool
 async def list_timelines(project_id: str = "") -> str:

app/api/middleware/auth.py

@@ -65,16 +65,39 @@ async def get_current_user(
     default_tier = "power" if settings.ENV == "dev" else "free"
     tier: str = result.scalar_one_or_none() or default_tier
 
-    # Fetch name/surname from user row.
+    # Fetch name/surname/avatar_url/onboarding_completed_at/password_hash from user row.
     user_result = await db.execute(
-        select(User.name, User.surname).where(User.id == user_id)
+        select(
+            User.name, User.surname, User.avatar_url, User.onboarding_completed_at,
+            User.password_hash,
+        ).where(User.id == user_id)
     )
     user_row = user_result.one_or_none()
 
+    # Convert onboarding_completed_at to epoch ms (int) or None.
+    onboarding_ms: int | None = None
+    if user_row and user_row.onboarding_completed_at is not None:
+        onboarding_ms = int(user_row.onboarding_completed_at.timestamp() * 1000)
+
+    # Load decrypted core memory.
+    from app.core.memory_middleware import MemoryMiddleware  # noqa: PLC0415
+
+    memory_dict: dict[str, str] = {}
+    try:
+        mw = MemoryMiddleware(db)
+        blocks = await mw.list_core_blocks(user_id)
+        memory_dict = {b["label"]: b["value"] for b in blocks}
+    except Exception:
+        pass  # Non-critical — return empty memory on failure
+
     return UserProfile(
         id=user_id,
         email=email,
         name=user_row.name if user_row else None,
         surname=user_row.surname if user_row else None,
+        avatar_url=user_row.avatar_url if user_row else None,
+        has_password=bool(user_row.password_hash) if user_row else False,
         tier=tier,
+        onboarding_completed_at=onboarding_ms,
+        memory=memory_dict,
     )  # type: ignore[arg-type]

app/api/middleware/sanitizer.py

@@ -8,8 +8,7 @@ that could reveal server-side prompt IP:
   - Internal reasoning markers (<thinking>, <reasoning>, [INST], …)
   - Exact-match known prompt fingerprints
 
-Binary responses (storage blobs, backup data) are never touched — the
+The middleware only activates for paths under /api/v1/chat.
-middleware only activates for paths under /api/v1/chat.
 
 Any sanitisation event is logged as a WARNING with the request path and the
 names of the fields that were modified.

app/api/routes/agent_setup.py

@@ -1,11 +1,11 @@
-"""Chatbot Journey — WS-based guided conversation to build an agent prompt_template.
+"""Chatbot Journey — WS-based guided conversation to build an AgentConfig.
 
 The journey is driven entirely through WebSocket frames (no REST endpoints).
 The device WS handler dispatches ``journey_start`` and ``journey_message``
 frames to the functions exported here.
 
 Journey flow:
-  1. FE sends ``journey_start`` frame with basic agent config (directory,
+  1. FE sends ``journey_start`` frame with basic agent info (directory,
      data_types, schedule).
   2. Server creates an in-memory session, sets up a WS executor so the
      setup LLM can use file-system tools, does a first directory scrape,
@@ -13,10 +13,11 @@ Journey flow:
   3. FE sends ``journey_message`` frames for each user reply.
   4. Server appends the user message, calls the LLM (which may read files
      via tools), and sends back a ``journey_reply``.
-  5. After 3-5 turns the LLM wraps up by emitting a ``prompt_template``
+  5. After 3-5 turns the LLM wraps up by emitting an ``AgentConfig`` JSON
-     block delimited by ``PROMPT_TEMPLATE_START`` / ``PROMPT_TEMPLATE_END``.
+     block delimited by ``AGENT_CONFIG_START`` / ``AGENT_CONFIG_END``.
-  6. Server parses the block, sends ``journey_reply`` with ``done=True``
+  6. Server parses and validates the JSON with Pydantic, sends
-     and the template.  FE stores it locally.
+     ``journey_reply`` with ``done=True`` and the serialised config.
+     FE stores it locally.
 """
 
 from __future__ import annotations
@@ -30,8 +31,10 @@ from typing import Any
 
 from langchain_core.messages import AIMessage, HumanMessage, SystemMessage, ToolMessage
 
-from app.agents.filesystem_agent import FILESYSTEM_TOOLS
+from app.agents.filesystem_agent import make_directory_tools
-from app.core.llm import get_llm
+from app.core.langfuse_client import compile_prompt, extract_usage, get_langfuse, get_prompt_or_fallback, langfuse_context
+from app.core.llm import get_agent_llm, model_for_agent
+from app.schemas import AgentConfig
 
 logger = logging.getLogger(__name__)
 
@@ -39,9 +42,9 @@ logger = logging.getLogger(__name__)
 
 _SESSION_TTL_SECONDS: int = 1800  # 30 minutes
 
-# Sentinel strings used to delimit the LLM-produced prompt_template.
+# Sentinel strings used to delimit the LLM-produced AgentConfig JSON.
-_TEMPLATE_START = "PROMPT_TEMPLATE_START"
+_CONFIG_START = "AGENT_CONFIG_START"
-_TEMPLATE_END = "PROMPT_TEMPLATE_END"
+_CONFIG_END = "AGENT_CONFIG_END"
 
 # Minimum turns before we consider nudging the LLM to wrap up.
 _MIN_TURNS_BEFORE_NUDGE: int = 3
@@ -62,6 +65,7 @@ class JourneySession:
     data_types: list[str]
     history: list[dict[str, Any]] = field(default_factory=list)
     system_prompt: str = ""
+    langfuse_prompt: Any = None
     created_at: float = field(default_factory=time.monotonic)
 
     def is_expired(self) -> bool:
@@ -83,61 +87,76 @@ def get_journey_session(session_id: str, user_id: str) -> JourneySession | None:
     return s
 
 
-# ── System prompt builder ─────────────────────────────────────────────────
+# ── System prompt ─────────────────────────────────────────────────────────
 
-_SYSTEM_PROMPT_TEMPLATE = """\
+_JOURNEY_SYSTEM_PROMPT = """\
 You are a friendly assistant helping a freelancer configure a data-extraction agent.
-Your job is to understand exactly what data the user wants to extract from their
+Your job is to understand what files the user has in their directory and produce a
-local directory and produce a detailed prompt_template that a separate AI will use
+structured AgentConfig JSON that the extraction agent will use as its instruction set.
-as its instruction set.
-
-The extraction agent already has this base behaviour built in:
-  - Reads each file using file-system tools.
-  - Creates records (tasks, notes, timelines, projects) via CRUD tools.
-  - Sets isAiSuggested=1 on every new record.
-  - Only extracts data explicitly present in the files — it never invents information.
-The user's custom prompt is appended AFTER this base behaviour, so focus on
-what to look for and how to map it — not on the general extraction mechanics.
 
 You have access to file-system tools to explore the user's directory:
-- list_directory: to see folder structure
+- list_directory: see folder structure and file names
-- read_file_content: to peek at file contents
+- read_file_content: peek at a file's content
-- get_file_metadata: to check file info
+- get_file_metadata: check file size, extension, dates
 
 The user's configured directory is: {directory}
 Target data types: {data_types}
 
-IMPORTANT — project assignment is handled automatically by the main agent runner
+## Your process
-before the custom prompt is ever used.  You MUST NOT ask the user about projects,
-projectId, or how to link records to projects.  Never include projectId logic or
-project creation instructions in the generated prompt_template.
 
-Start by exploring the directory to understand its structure.  Then ask concise,
+### Step 1 — Explore the directory
-focused questions one at a time.  Cover these topics (not necessarily in this order):
+Use list_directory and read_file_content to understand what types of files are present
-  1. The type and format of the source content (confirmed by your exploration).
+(HTML emails, plain-text documents, CSVs, etc.).
-  2. How fields should be mapped (e.g. filename → task title).
-  3. Priority or status rules (e.g. "urgent" keyword → high priority).
-  4. Any special handling, date extraction, or exclusions.
 
-Once you reach 90% confidence, output the final prompt_template between these exact
+### Step 2 — Identify content types
-markers on their own lines:
+For each distinct file type found, decide:
+- A short id (e.g. "email_html", "plain_text", "csv")
+- Which preprocessing handler to use: "email_html" for HTML emails, "generic" for everything else
+- A human-readable label and optional detection_hint
 
-{template_start}
+### Step 3 — Ask focused questions (one at a time)
-<the complete extraction prompt here>
+Cover these topics based on what you discovered:
-{template_end}
+1. How to map content to entity types (task / note / timeline entry)
+2. Field mapping rules (e.g. email Subject → task title, filename → note title)
+3. Priority or status rules (e.g. "urgent" in subject → high priority)
+4. Date extraction (e.g. "by Friday" → dueDate)
+5. Exclusion rules (e.g. skip newsletters, skip files with no project match)
 
-The prompt_template must be a self-contained instruction for an AI that reads files
+### Step 4 — Produce the AgentConfig JSON
-and must perform CRUD operations using tools to create records.  It should specify:
+Once you are ≥ 90% confident, output the final config between these exact markers
-  - What entity types to create (tasks, notes, timelines) — never projects.
+(each on its own line):
-  - How to map file content to record fields (camelCase: title, status, priority,
+
-    dueDate, content, etc.) — never include projectId.
+{config_start}
-  - That isAiSuggested must be set to 1 on every new record.
+{{
-  - Concrete examples of mappings based on what you discovered in the directory.
+  "content_types": [
+    {{
+      "id": "email_html",
+      "label": "Email HTML",
+      "detection_hint": "HTML file with From/To/Subject headers",
+      "preprocessing": "email_html",
+      "extraction_prompt": "Detailed extraction instructions for this content type..."
+    }}
+  ],
+  "global_rules": [
+    "If the file cannot be matched to any project, do not create any entity."
+  ],
+  "data_types": {data_types_json}
+}}
+{config_end}
+
+## Rules for the extraction_prompt field
+- Describe when to create a task vs note vs timeline entry (be specific and concrete)
+- Include field mapping rules based on what you found in the directory
+- Include priority/status/date rules if applicable
+- Do NOT include projectId logic — the runner handles project assignment automatically
+- Do NOT mention isAiSuggested — the runner always sets it to 1
+
+## Constraints
+- Never ask about projects, projectId, or how to link records to projects
+- Never include projectId or project creation logic in the generated config
+- Keep asking questions until ≥ 90% confident, then output the JSON immediately
 
 {existing_section}\
-Keep asking clarifying questions until you are at least 90% confident you have
-enough information to generate an accurate prompt_template.  Once you reach that
-confidence level, stop asking and produce the final template immediately.
 Begin by exploring the directory, then ask your first question.\
 """
 
@@ -145,33 +164,53 @@ Begin by exploring the directory, then ask your first question.\
 def _build_system_prompt(
     directory: str,
     data_types: list[str],
-    existing_template: str | None = None,
+    existing_config: str | None = None,
-) -> str:
+) -> tuple[str, Any]:
+    """Return ``(compiled_system_prompt, langfuse_prompt_obj_or_None)``."""
     existing_section = (
-        f"\nThe user already has the following prompt_template — refine it based on their answers:\n"
+        "\nThe user already has the following AgentConfig — refine it based on their answers:\n"
-        f"---\n{existing_template}\n---\n"
+        f"```json\n{existing_config}\n```\n"
-        if existing_template
+        if existing_config
         else ""
     )
-    return _SYSTEM_PROMPT_TEMPLATE.format(
+    template, prompt_obj = get_prompt_or_fallback(
+        "journey_system", _JOURNEY_SYSTEM_PROMPT
+    )
+    compiled = compile_prompt(
+        template,
+        prompt_obj,
         directory=directory,
         data_types=", ".join(data_types),
-        template_start=_TEMPLATE_START,
+        data_types_json=json.dumps(data_types),
-        template_end=_TEMPLATE_END,
+        config_start=_CONFIG_START,
+        config_end=_CONFIG_END,
         existing_section=existing_section,
     )
+    return compiled, prompt_obj
 
 
-# ── Template extraction ───────────────────────────────────────────────────
+# ── AgentConfig extraction ────────────────────────────────────────────────
 
 
-def _extract_template(text: str) -> str | None:
+def _extract_agent_config(text: str) -> str | None:
-    """Return the text between PROMPT_TEMPLATE_START and PROMPT_TEMPLATE_END, or None."""
+    """Return validated AgentConfig JSON string from between markers, or None.
-    if _TEMPLATE_START not in text or _TEMPLATE_END not in text:
+
+    Parses the JSON with Pydantic to ensure it conforms to the schema before
+    returning.  Returns None if markers are absent or JSON is invalid.
+    """
+    if _CONFIG_START not in text or _CONFIG_END not in text:
+        return None
+    start_idx = text.index(_CONFIG_START) + len(_CONFIG_START)
+    end_idx = text.index(_CONFIG_END)
+    raw = text[start_idx:end_idx].strip()
+    if not raw:
+        return None
+    try:
+        parsed = AgentConfig.model_validate_json(raw)
+        return parsed.model_dump_json()
+    except Exception as exc:
+        logger.warning("agent_setup: failed to parse AgentConfig JSON: %s", exc)
         return None
-    start_idx = text.index(_TEMPLATE_START) + len(_TEMPLATE_START)
-    end_idx = text.index(_TEMPLATE_END)
-    return text[start_idx:end_idx].strip() or None
 
 
 # ── LLM call with tool support ───────────────────────────────────────────
@@ -199,12 +238,17 @@ async def _call_llm_with_tools(
     system_prompt: str,
     history: list[dict[str, Any]],
     tools: list[Any],
+    *,
+    user_id: str = "",
+    session_id: str = "",
+    langfuse_prompt: Any = None,
 ) -> str:
     """Build LangChain messages from history and invoke the LLM with tools.
 
     Handles tool-calling loops: if the LLM calls tools, execute them and
     continue until a final text response is produced.
     """
+    lf = get_langfuse()
     messages: list[Any] = [SystemMessage(content=system_prompt)]
     for turn in history:
         if turn["role"] == "user":
@@ -212,42 +256,97 @@ async def _call_llm_with_tools(
         else:
             messages.append(AIMessage(content=turn["content"]))
 
-    llm = get_llm(model=None, temperature=0.4)
+    llm = get_agent_llm("setup", temperature=0.4)
     llm_with_tools = llm.bind_tools(tools)
     tool_map = {tool_def.name: tool_def for tool_def in tools}
 
-    for _ in range(_MAX_TOOL_STEPS):
+    _lf_ctx = langfuse_context(user_id=user_id or None, session_id=session_id or None)
-        response: AIMessage = await llm_with_tools.ainvoke(messages)
+    _lf_ctx.__enter__()
-        messages.append(response)
 
-        if not response.tool_calls:
+    _span_ctx = (
-            return _as_text(response.content)
+        lf.start_as_current_observation(
+            as_type="span",
+            name="journey-setup",
+            input=history[-1]["content"] if history else "",
+        )
+        if lf else None
+    )
+    _span = _span_ctx.__enter__() if _span_ctx else None
 
-        for call in response.tool_calls:
+    try:
-            call_name = str(call.get("name", ""))
+        for step in range(_MAX_TOOL_STEPS):
-            call_args = call.get("args", {})
+            _gen_ctx = (
-            logger.info(
+                lf.start_as_current_observation(
-                "agent_setup: journey tool_call name=%s args=%s",
+                    as_type="generation",
-                call_name,
+                    name="journey-setup-llm",
-                json.dumps(call_args, ensure_ascii=True)[:500],
+                    model=model_for_agent("setup"),
+                    prompt=langfuse_prompt,
+                    input=messages,
+                )
+                if lf else None
             )
+            _gen = _gen_ctx.__enter__() if _gen_ctx else None
+            response: AIMessage = await llm_with_tools.ainvoke(messages)
+            if _gen_ctx:
+                _gen.update(output=_as_text(response.content), usage_details=extract_usage(response))
+                _gen_ctx.__exit__(None, None, None)
 
-            tool_fn = tool_map.get(call_name)
+            resp_text = _as_text(response.content)
-            if tool_fn is None:
-                tool_output = f"Unknown tool: {call_name}"
-            else:
-                tool_output = await tool_fn.ainvoke(call_args)
 
-            logger.info(
+            # Guard against empty responses (e.g. model returned finish_reason
-                "agent_setup: journey tool_result name=%s output=%s",
+            # 'error' which LiteLLM maps to 'stop' with empty content).
-                call_name,
+            if not response.tool_calls and not resp_text.strip():
-                str(tool_output)[:800],
+                logger.warning(
-            )
+                    "agent_setup: journey LLM returned empty response at step %d — retrying",
-            messages.append(ToolMessage(content=str(tool_output), tool_call_id=call["id"]))
+                    step,
+                )
+                # Drop the empty AIMessage so we don't pollute history, and retry.
+                continue
 
-    # Fallback: exceeded max steps.
+            messages.append(response)
-    final = await llm.ainvoke(messages)
+
-    return _as_text(final.content)
+            if not response.tool_calls:
+                if _span:
+                    _span.update(output=resp_text)
+                return resp_text
+
+            for call in response.tool_calls:
+                call_name = str(call.get("name", ""))
+                call_args = call.get("args", {})
+                logger.info(
+                    "agent_setup: journey tool_call name=%s args=%s",
+                    call_name,
+                    json.dumps(call_args, ensure_ascii=True)[:500],
+                )
+
+                tool_fn = tool_map.get(call_name)
+                if tool_fn is None:
+                    tool_output = f"Unknown tool: {call_name}"
+                else:
+                    tool_output = await tool_fn.ainvoke(call_args)
+
+                logger.info(
+                    "agent_setup: journey tool_result name=%s output=%s",
+                    call_name,
+                    str(tool_output)[:800],
+                )
+                messages.append(ToolMessage(content=str(tool_output), tool_call_id=call["id"]))
+
+        # Fallback: exceeded max steps.
+        final = await llm.ainvoke(messages)
+        final_text = _as_text(final.content)
+        if _span:
+            _span.update(output=final_text)
+        return final_text or (
+            "Sorry, I had trouble processing the files. "
+            "Could you try again? If the issue persists, the files might be too large for me to analyse."
+        )
+    finally:
+        if _span_ctx:
+            _span_ctx.__exit__(None, None, None)
+        _lf_ctx.__exit__(None, None, None)
+        if lf:
+            lf.flush()
 
 
 # ── Journey handlers (called from device_ws.py) ──────────────────────────
@@ -265,12 +364,12 @@ async def handle_journey_start(
     agent_type = frame.get("agent_type", "local")
     directory = frame.get("directory", "")
     data_types = frame.get("data_types", [])
-    existing_template = frame.get("existing_template")
+    existing_config = frame.get("existing_config")
 
     # Use the session_id provided by the FE so the reply matches the
     # listener key; fall back to a generated one if absent.
     session_id = frame.get("session_id") or str(uuid.uuid4())
-    system_prompt = _build_system_prompt(directory, data_types, existing_template)
+    system_prompt, langfuse_prompt = _build_system_prompt(directory, data_types, existing_config)
 
     session = JourneySession(
         session_id=session_id,
@@ -279,19 +378,21 @@ async def handle_journey_start(
         directory=directory,
         data_types=data_types,
         system_prompt=system_prompt,
+        langfuse_prompt=langfuse_prompt,
     )
 
-    # The LLM will explore the directory using FILESYSTEM_TOOLS via the
+    # Seed with an initial user message — some providers require at least one
-    # ws_context executor (already set by the WS handler before calling us).
+    # user/input message to be present.
-    # Seed with an initial user message — some providers (e.g. GitHub Copilot)
-    # require at least one user/input message to be present.
     seed_history: list[dict[str, Any]] = [
         {"role": "user", "content": "Hi, I'm ready to set up my agent. Please explore my directory and ask me your first question."},
     ]
     ai_reply = await _call_llm_with_tools(
         system_prompt=system_prompt,
         history=seed_history,
-        tools=list(FILESYSTEM_TOOLS),
+        tools=make_directory_tools(directory),
+        user_id=user_id,
+        session_id=session_id,
+        langfuse_prompt=langfuse_prompt,
     )
 
     session.history.extend(seed_history)
@@ -305,14 +406,14 @@ async def handle_journey_start(
         directory,
     )
 
-    # Check if the LLM produced the template on the first turn (unlikely but possible).
+    # Check if the LLM produced the config on the first turn (unlikely but possible).
-    prompt_template = _extract_template(ai_reply)
+    agent_config = _extract_agent_config(ai_reply)
-    done = prompt_template is not None
+    done = agent_config is not None
 
     display_message = ai_reply
     if done:
         display_message = (
-            ai_reply[: ai_reply.index(_TEMPLATE_START)].strip()
+            ai_reply[: ai_reply.index(_CONFIG_START)].strip()
             or "Here is your agent configuration. You can save it or continue refining."
         )
         _sessions.pop(session_id, None)
@@ -322,7 +423,7 @@ async def handle_journey_start(
         "session_id": session_id,
         "message": display_message,
         "done": done,
-        "prompt_template": prompt_template,
+        "agent_config": agent_config,
     }
 
 
@@ -345,53 +446,59 @@ async def handle_journey_message(
             "session_id": session_id,
             "message": "Journey session not found or expired. Please start a new setup.",
             "done": True,
-            "prompt_template": None,
+            "agent_config": None,
         }
 
     # Append user turn.
     session.history.append({"role": "user", "content": message})
 
     # Call the LLM with tools.
+    session_tools = make_directory_tools(session.directory)
     ai_reply = await _call_llm_with_tools(
         system_prompt=session.system_prompt,
         history=session.history,
-        tools=list(FILESYSTEM_TOOLS),
+        tools=session_tools,
+        user_id=session.user_id,
+        session_id=session_id,
+        langfuse_prompt=session.langfuse_prompt,
     )
 
     session.history.append({"role": "assistant", "content": ai_reply})
 
-    # Check if the LLM produced the final template.
+    # Check if the LLM produced the final config.
-    prompt_template = _extract_template(ai_reply)
+    agent_config = _extract_agent_config(ai_reply)
-    done = prompt_template is not None
+    done = agent_config is not None
 
-    # If the LLM didn't produce a template, nudge it once it has asked enough
+    # If the LLM didn't produce a config, nudge it once it hits the hard safety cap.
-    # questions (>= _MIN_TURNS_BEFORE_NUDGE) or hits the hard safety cap.
     if not done:
         turns = sum(1 for t in session.history if t["role"] == "user")
         if turns >= _MAX_TURNS:
             nudge_content = (
                 "[System: You have enough information. Please generate the final "
-                f"prompt_template now, wrapped in {_TEMPLATE_START} / {_TEMPLATE_END} markers.]"
+                f"AgentConfig JSON now, wrapped in {_CONFIG_START} / {_CONFIG_END} markers.]"
             )
             session.history.append({"role": "user", "content": nudge_content})
 
             nudge_reply = await _call_llm_with_tools(
                 system_prompt=session.system_prompt,
                 history=session.history,
-                tools=list(FILESYSTEM_TOOLS),
+                tools=session_tools,
+                user_id=session.user_id,
+                session_id=session_id,
+                langfuse_prompt=session.langfuse_prompt,
             )
             session.history.append({"role": "assistant", "content": nudge_reply})
 
-            prompt_template = _extract_template(nudge_reply)
+            agent_config = _extract_agent_config(nudge_reply)
-            if prompt_template is not None:
+            if agent_config is not None:
                 done = True
                 ai_reply = nudge_reply
 
     display_message = ai_reply
     if done:
         display_message = (
-            ai_reply[: ai_reply.index(_TEMPLATE_START)].strip()
+            ai_reply[: ai_reply.index(_CONFIG_START)].strip()
-            if _TEMPLATE_START in ai_reply
+            if _CONFIG_START in ai_reply
             else "Here is your agent configuration. You can save it or continue refining."
         )
         _sessions.pop(session_id, None)
@@ -402,5 +509,5 @@ async def handle_journey_message(
         "session_id": session_id,
         "message": display_message,
         "done": done,
-        "prompt_template": prompt_template,
+        "agent_config": agent_config,
     }

app/api/routes/agents.py

@@ -12,8 +12,11 @@ in backend agent-config tables.
 from __future__ import annotations
 
 import asyncio
+import logging
 import uuid
-from datetime import datetime, timedelta, timezone
+from datetime import datetime, timezone
+
+logger = logging.getLogger(__name__)
 
 from fastapi import APIRouter, Depends, HTTPException, status
 from sqlalchemy import func, select
@@ -177,6 +180,11 @@ async def trigger_agent_run(
     _enforce_agent_limit(current_user.tier, body.active_agents)
     await _enforce_run_frequency(current_user.tier, current_user.id, db)
 
+    last_run_dt = (
+        datetime.fromtimestamp(body.last_run_at / 1000, tz=timezone.utc)
+        if body.last_run_at
+        else None
+    )
     config = LocalAgentConfig(
         id=str(uuid.uuid4()),
         user_id=current_user.id,
@@ -184,10 +192,12 @@ async def trigger_agent_run(
         name="Local Directory Monitor",
         directory_paths=[body.directory],
         data_types=_to_data_types(body.what_to_extract),
-        prompt_template=body.custom_agent_prompt,
+        prompt_template=body.custom_agent_prompt or "",
+        agent_config=body.agent_config,
         file_extensions=[],
         schedule_cron=body.batch_interval,
         enabled=True,
+        last_run_at=last_run_dt,
     )
 
     # Use the FE's stable agent_id if provided, fall back to the ephemeral config id.

app/api/routes/auth.py

@@ -1,34 +1,68 @@
-"""Auth routes: register, login, refresh, me.
+"""Auth routes: register, login, refresh, me, OAuth social login, onboarding.
 
 Users and refresh tokens are persisted in PostgreSQL (users + refresh_tokens
 tables).  Passwords are hashed with bcrypt; refresh tokens are stored as
 SHA-256 hashes so plaintext never reaches the DB.
+
+OAuth (Google):
+  GET  /auth/oauth/{provider}/authorize  — returns consent-screen URL + state
+  POST /auth/oauth/{provider}/callback   — exchanges code, issues JWT tokens
 """
 
 from __future__ import annotations
 
 import hashlib
+import json
 import time
+import urllib.parse
 import uuid
 from datetime import datetime, timedelta, timezone
+from typing import Literal
 
 import bcrypt
 from cryptography.fernet import Fernet
 from fastapi import APIRouter, Depends, HTTPException, status
+from fastapi.responses import RedirectResponse
 from jose import jwt
-from pydantic import BaseModel
+from pydantic import BaseModel, Field
 from sqlalchemy import select
 from sqlalchemy.ext.asyncio import AsyncSession
 
 from app.api.deps import get_current_user
+from app.auth.oauth_providers import GoogleOAuthProvider, generate_pkce_pair
 from app.config.settings import settings
+from app.core.llm import get_llm
+from app.core.memory_middleware import MemoryMiddleware
 from app.db import get_session
-from app.models import RefreshToken, User
+from app.models import OAuthAccount, RefreshToken, User
 from app.schemas import AuthTokens, UserProfile
 
 router = APIRouter(prefix="/auth", tags=["auth"])
 
 
+# ── OAuth provider registry ───────────────────────────────────────────
+
+def _get_google_provider() -> GoogleOAuthProvider:
+    if not settings.GOOGLE_AUTH_CLIENT_ID or not settings.GOOGLE_AUTH_CLIENT_SECRET:
+        raise HTTPException(
+            status.HTTP_503_SERVICE_UNAVAILABLE,
+            "Google login is not configured on this server",
+        )
+    return GoogleOAuthProvider(
+        client_id=settings.GOOGLE_AUTH_CLIENT_ID,
+        client_secret=settings.GOOGLE_AUTH_CLIENT_SECRET,
+        redirect_uri=settings.OAUTH_REDIRECT_URI,
+    )
+
+
+_PROVIDERS = {"google": _get_google_provider}
+
+# In-memory state store: state → (code_verifier, expires_at_epoch_s)
+# Production note: replace with Redis for multi-process deployments.
+_pending_states: dict[str, tuple[str, float]] = {}
+_STATE_TTL_SECONDS = 600  # 10 minutes
+
+
 # ── Internal helpers ─────────────────────────────────────────────────
 
 
@@ -231,5 +265,531 @@ async def update_profile(
         email=user.email,
         name=user.name,
         surname=user.surname,
+        avatar_url=user.avatar_url,
         tier=current_user.tier,
     )
+
+
+# ── OAuth helpers ─────────────────────────────────────────────────────
+
+
+async def _issue_refresh_token(user: User, db: AsyncSession) -> tuple[str, AuthTokens]:
+    """Create a refresh token row and return (plain_token, AuthTokens)."""
+    plain_token = str(uuid.uuid4())
+    expires_at = datetime.now(timezone.utc) + timedelta(
+        days=settings.JWT_REFRESH_TOKEN_EXPIRE_DAYS
+    )
+    rt = RefreshToken(
+        user_id=user.id,
+        token_hash=_hash_token(plain_token),
+        expires_at=expires_at,
+    )
+    db.add(rt)
+    access_token, expires_at_ms = _make_access_token(user.id, user.email, user.tier)
+    return plain_token, AuthTokens(
+        access_token=access_token,
+        refresh_token=plain_token,
+        expires_at=expires_at_ms,
+    )
+
+
+# ── OAuth request/response schemas ───────────────────────────────────
+
+
+class _OAuthAuthorizeResponse(BaseModel):
+    url: str
+    state: str
+
+
+class _OAuthCallbackRequest(BaseModel):
+    code: str
+    state: str
+
+
+# ── OAuth routes ──────────────────────────────────────────────────────
+
+
+@router.get(
+    "/oauth/{provider}/web-callback",
+    summary="Web-facing OAuth redirect — bounces to the adiuvai:// deep link",
+    include_in_schema=False,
+)
+async def oauth_web_callback(
+    provider: Literal["google"],
+    code: str,
+    state: str,
+) -> RedirectResponse:
+    """Google redirects here after user consent.
+
+    This endpoint immediately redirects to the Electron deep-link URI so the
+    desktop app receives the authorization code.  It is intentionally simple —
+    no state validation here (the Electron app + backend callback do that).
+
+    Registered in Google Cloud Console as:
+      http://localhost:8000/api/v1/auth/oauth/google/web-callback  (dev)
+      https://api.adiuvai.com/api/v1/auth/oauth/google/web-callback  (prod)
+    """
+    params = urllib.parse.urlencode({"code": code, "state": state, "provider": provider})
+    deep_link = f"adiuvai://oauth/callback?{params}"
+    return RedirectResponse(url=deep_link, status_code=302)
+
+
+@router.get(
+    "/oauth/{provider}/authorize",
+    response_model=_OAuthAuthorizeResponse,
+    summary="Start OAuth flow — returns the provider consent-screen URL",
+)
+async def oauth_authorize(
+    provider: Literal["google"],
+) -> _OAuthAuthorizeResponse:
+    """Generate a PKCE state + code_challenge and return the authorization URL.
+
+    The client opens this URL in the system browser.  After the user grants
+    consent, the provider redirects to the deep-link URI (adiuvai://oauth/callback)
+    with ``code`` and ``state`` query params.  The client then calls
+    ``POST /auth/oauth/{provider}/callback`` with those values.
+    """
+    provider_factory = _PROVIDERS.get(provider)
+    if provider_factory is None:
+        raise HTTPException(status.HTTP_400_BAD_REQUEST, f"Unknown provider: {provider}")
+
+    oauth_provider = provider_factory()
+    state = str(uuid.uuid4())
+    code_verifier, code_challenge = generate_pkce_pair()
+
+    # Purge expired states to prevent unbounded growth.
+    now = time.time()
+    expired = [s for s, (_, exp) in _pending_states.items() if exp < now]
+    for s in expired:
+        del _pending_states[s]
+
+    _pending_states[state] = (code_verifier, now + _STATE_TTL_SECONDS)
+
+    url = oauth_provider.get_authorization_url(state=state, code_challenge=code_challenge)
+    return _OAuthAuthorizeResponse(url=url, state=state)
+
+
+@router.post(
+    "/oauth/{provider}/callback",
+    response_model=AuthTokens,
+    summary="Complete OAuth flow — exchange code and issue JWT tokens",
+)
+async def oauth_callback(
+    provider: Literal["google"],
+    body: _OAuthCallbackRequest,
+    db: AsyncSession = Depends(get_session),
+) -> AuthTokens:
+    """Validate state, exchange the authorization code, and sign in (or register) the user.
+
+    Resolution order:
+      1. ``oauth_accounts`` row match → existing user, log in.
+      2. Email match + ``email_verified=True`` → link OAuth account to existing user.
+      3. No match → create new user (password_hash=None, avatar from provider).
+    """
+    provider_factory = _PROVIDERS.get(provider)
+    if provider_factory is None:
+        raise HTTPException(status.HTTP_400_BAD_REQUEST, f"Unknown provider: {provider}")
+
+    # Validate state (CSRF protection).
+    now = time.time()
+    entry = _pending_states.pop(body.state, None)
+    if entry is None or entry[1] < now:
+        raise HTTPException(status.HTTP_401_UNAUTHORIZED, "Invalid or expired OAuth state")
+
+    code_verifier, _ = entry
+
+    oauth_provider = provider_factory()
+
+    # Exchange code for tokens.
+    try:
+        token_data = await oauth_provider.exchange_code(
+            code=body.code,
+            code_verifier=code_verifier,
+            redirect_uri=settings.OAUTH_REDIRECT_URI,
+        )
+    except Exception:
+        raise HTTPException(
+            status.HTTP_400_BAD_REQUEST, "Failed to exchange authorization code"
+        )
+
+    access_token_google = token_data.get("access_token")
+    if not access_token_google:
+        raise HTTPException(status.HTTP_400_BAD_REQUEST, "No access token in provider response")
+
+    # Fetch user identity.
+    try:
+        userinfo = await oauth_provider.get_userinfo(access_token_google)
+    except Exception:
+        raise HTTPException(status.HTTP_400_BAD_REQUEST, "Failed to fetch user info from provider")
+
+    # ── Resolution order ──────────────────────────────────────────────
+
+    # 1. Existing OAuth link?
+    oauth_result = await db.execute(
+        select(OAuthAccount).where(
+            OAuthAccount.provider == provider,
+            OAuthAccount.provider_user_id == userinfo.provider_user_id,
+        )
+    )
+    oauth_account = oauth_result.scalar_one_or_none()
+
+    if oauth_account is not None:
+        user_result = await db.execute(select(User).where(User.id == oauth_account.user_id))
+        user = user_result.scalar_one()
+        # Backfill avatar if the user doesn't have one yet.
+        if user.avatar_url is None and userinfo.avatar_url:
+            user.avatar_url = userinfo.avatar_url
+            await db.commit()
+        plain_token, tokens = await _issue_refresh_token(user, db)
+        await db.commit()
+        return tokens
+
+    # 2. Email match with a verified Google email → link accounts.
+    if userinfo.email_verified:
+        email_result = await db.execute(select(User).where(User.email == userinfo.email))
+        existing_user = email_result.scalar_one_or_none()
+
+        if existing_user is not None:
+            new_link = OAuthAccount(
+                user_id=existing_user.id,
+                provider=provider,
+                provider_user_id=userinfo.provider_user_id,
+                provider_email=userinfo.email,
+            )
+            db.add(new_link)
+            if existing_user.avatar_url is None and userinfo.avatar_url:
+                existing_user.avatar_url = userinfo.avatar_url
+            plain_token, tokens = await _issue_refresh_token(existing_user, db)
+            await db.commit()
+            return tokens
+
+    # Guard: if the email is already taken but we couldn't auto-link (e.g.
+    # email_verified=False), refuse with 409 instead of hitting a DB constraint.
+    if not userinfo.email_verified:
+        conflict = await db.execute(select(User).where(User.email == userinfo.email))
+        if conflict.scalar_one_or_none() is not None:
+            raise HTTPException(
+                status.HTTP_409_CONFLICT,
+                "An account with this email already exists. "
+                "Please sign in with your password.",
+            )
+
+    # 3. New user — social-only account (no password).
+    new_user = User(
+        id=str(uuid.uuid4()),
+        email=userinfo.email,
+        name=userinfo.name,
+        password_hash=None,
+        avatar_url=userinfo.avatar_url,
+        tier="free",
+        encryption_key=Fernet.generate_key().decode(),
+    )
+    db.add(new_user)
+    await db.flush()  # populate new_user.id
+
+    new_oauth = OAuthAccount(
+        user_id=new_user.id,
+        provider=provider,
+        provider_user_id=userinfo.provider_user_id,
+        provider_email=userinfo.email,
+    )
+    db.add(new_oauth)
+
+    plain_token, tokens = await _issue_refresh_token(new_user, db)
+    await db.commit()
+    return tokens
+
+
+# ── Onboarding helpers ────────────────────────────────────────────────
+
+
+async def _build_profile(user_id: str, email: str, db: AsyncSession) -> UserProfile:
+    """Re-fetch and return a full UserProfile (reuses get_current_user logic)."""
+
+    # We can't call the FastAPI dependency directly, but we can replicate
+    # the core logic inline.  Instead, we just re-query the same way.
+    from app.models import Subscription  # noqa: PLC0415
+
+    result = await db.execute(
+        select(Subscription.tier).where(Subscription.user_id == user_id)
+    )
+    default_tier = "power" if settings.ENV == "dev" else "free"
+    tier: str = result.scalar_one_or_none() or default_tier
+
+    user_result = await db.execute(
+        select(
+            User.name, User.surname, User.avatar_url, User.onboarding_completed_at,
+            User.password_hash,
+        ).where(User.id == user_id)
+    )
+    user_row = user_result.one_or_none()
+
+    onboarding_ms: int | None = None
+    if user_row and user_row.onboarding_completed_at is not None:
+        onboarding_ms = int(user_row.onboarding_completed_at.timestamp() * 1000)
+
+    memory_dict: dict[str, str] = {}
+    try:
+        mw = MemoryMiddleware(db)
+        blocks = await mw.list_core_blocks(user_id)
+        memory_dict = {b["label"]: b["value"] for b in blocks}
+    except Exception:
+        pass
+
+    return UserProfile(
+        id=user_id,
+        email=email,
+        name=user_row.name if user_row else None,
+        surname=user_row.surname if user_row else None,
+        avatar_url=user_row.avatar_url if user_row else None,
+        has_password=bool(user_row.password_hash) if user_row else False,
+        tier=tier,
+        onboarding_completed_at=onboarding_ms,
+        memory=memory_dict,
+    )
+
+
+# ── Onboarding routes ────────────────────────────────────────────────
+
+
+class _UpdateMemoryRequest(BaseModel):
+    memory: dict[str, str] = Field(default_factory=dict)
+    mark_onboarded: bool = False
+
+
+@router.put("/me/memory", response_model=UserProfile)
+async def update_memory(
+    body: _UpdateMemoryRequest,
+    current_user: UserProfile = Depends(get_current_user),
+    db: AsyncSession = Depends(get_session),
+) -> UserProfile:
+    """Update core memory key/value pairs and optionally mark onboarding complete."""
+    mw = MemoryMiddleware(db)
+    for key, value in body.memory.items():
+        await mw.update_core(current_user.id, key, value)
+    if body.mark_onboarded:
+        result = await db.execute(select(User).where(User.id == current_user.id))
+        user = result.scalar_one()
+        user.onboarding_completed_at = datetime.now(timezone.utc)
+        await db.commit()
+    return await _build_profile(current_user.id, current_user.email, db)
+
+
+@router.post("/me/onboarding/reset")
+async def reset_onboarding(
+    current_user: UserProfile = Depends(get_current_user),
+    db: AsyncSession = Depends(get_session),
+):
+    """Reset onboarding so the wizard runs again on next login."""
+    result = await db.execute(select(User).where(User.id == current_user.id))
+    user = result.scalar_one()
+    user.onboarding_completed_at = None
+    await db.commit()
+    return {"status": "reset"}
+
+
+class _NormalizeRequest(BaseModel):
+    inputs: dict[str, str]
+
+
+class _NormalizeResponse(BaseModel):
+    normalized: dict[str, str]
+
+
+@router.post("/onboarding/normalize", response_model=_NormalizeResponse)
+async def normalize_onboarding(
+    body: _NormalizeRequest,
+    current_user: UserProfile = Depends(get_current_user),
+) -> _NormalizeResponse:
+    """One-shot LLM normalization for free-text onboarding answers."""
+    if not body.inputs:
+        return _NormalizeResponse(normalized={})
+    try:
+        llm = get_llm(model="gpt-4o-mini", temperature=0)
+        prompt = (
+            "You normalize user onboarding answers into clean, ≤3-word canonical labels.\n"
+            "Return a JSON object with the same keys and normalized values.\n"
+            "Examples: 'i build websites' → 'Web Developer', 'tech-ish stuff' → 'Technology'\n"
+            f"Input: {json.dumps(body.inputs)}"
+        )
+        response = await llm.ainvoke(
+            [
+                {"role": "system", "content": "You normalize user inputs. Return JSON only."},
+                {"role": "user", "content": prompt},
+            ],
+        )
+        normalized = json.loads(response.content)
+        return _NormalizeResponse(normalized=normalized)
+    except Exception:
+        # LLM failure must never block onboarding — return inputs unchanged
+        return _NormalizeResponse(normalized=body.inputs)
+
+
+# ── Password management ───────────────────────────────────────────────
+
+
+class _ChangePasswordRequest(BaseModel):
+    current_password: str = Field(min_length=1)
+    new_password: str = Field(min_length=8)
+
+
+@router.put("/me/password", status_code=status.HTTP_200_OK)
+async def change_password(
+    body: _ChangePasswordRequest,
+    current_user: UserProfile = Depends(get_current_user),
+    db: AsyncSession = Depends(get_session),
+) -> dict[str, bool]:
+    """Change the authenticated user's password.
+
+    Requires the current password for verification.
+    Returns 400 for social-only users (no password set).
+    """
+    result = await db.execute(select(User).where(User.id == current_user.id))
+    user = result.scalar_one()
+
+    if user.password_hash is None:
+        raise HTTPException(
+            status.HTTP_400_BAD_REQUEST,
+            "This account uses social login and has no password to change",
+        )
+
+    if not _verify_password(body.current_password, user.password_hash):
+        raise HTTPException(status.HTTP_400_BAD_REQUEST, "Current password is incorrect")
+
+    user.password_hash = _hash_password(body.new_password)
+    await db.commit()
+    return {"ok": True}
+
+
+# ── OAuth account management ─────────────────────────────────────────
+
+
+@router.get("/me/oauth-accounts", response_model=list[dict])
+async def list_oauth_accounts(
+    current_user: UserProfile = Depends(get_current_user),
+    db: AsyncSession = Depends(get_session),
+) -> list[dict]:
+    """List all OAuth providers linked to the authenticated user."""
+    result = await db.execute(
+        select(OAuthAccount).where(OAuthAccount.user_id == current_user.id)
+    )
+    accounts = result.scalars().all()
+    return [
+        {
+            "provider": a.provider,
+            "provider_email": a.provider_email,
+            "created_at": int(a.created_at.timestamp() * 1000),
+        }
+        for a in accounts
+    ]
+
+
+@router.delete("/me/oauth-accounts/{provider}", status_code=status.HTTP_200_OK)
+async def unlink_oauth_account(
+    provider: str,
+    current_user: UserProfile = Depends(get_current_user),
+    db: AsyncSession = Depends(get_session),
+) -> dict[str, bool]:
+    """Unlink an OAuth provider from the authenticated user.
+
+    Refuses if the user has no password and this is their only login method.
+    """
+    result = await db.execute(select(User).where(User.id == current_user.id))
+    user = result.scalar_one()
+
+    oauth_result = await db.execute(
+        select(OAuthAccount).where(
+            OAuthAccount.user_id == current_user.id,
+            OAuthAccount.provider == provider,
+        )
+    )
+    account = oauth_result.scalar_one_or_none()
+    if account is None:
+        raise HTTPException(status.HTTP_404_NOT_FOUND, f"No linked {provider} account found")
+
+    # Safety: don't let users lock themselves out.
+    all_oauth = await db.execute(
+        select(OAuthAccount).where(OAuthAccount.user_id == current_user.id)
+    )
+    oauth_count = len(all_oauth.scalars().all())
+
+    if user.password_hash is None and oauth_count <= 1:
+        raise HTTPException(
+            status.HTTP_400_BAD_REQUEST,
+            "Cannot unlink the only login method. Set a password first.",
+        )
+
+    await db.delete(account)
+    await db.commit()
+    return {"ok": True}
+
+
+# ── Avatar update ─────────────────────────────────────────────────────
+
+
+class _UpdateAvatarRequest(BaseModel):
+    avatar_url: str = Field(min_length=1)
+
+
+@router.put("/me/avatar", response_model=UserProfile)
+async def update_avatar(
+    body: _UpdateAvatarRequest,
+    current_user: UserProfile = Depends(get_current_user),
+    db: AsyncSession = Depends(get_session),
+) -> UserProfile:
+    """Update the authenticated user's avatar URL.
+
+    Accepts {"avatar_url": "https://..."} — the client uploads the image
+    to its own storage and passes the resulting URL here.
+    """
+    if not body.avatar_url.startswith(("https://", "http://", "data:image/")):
+        raise HTTPException(status.HTTP_400_BAD_REQUEST, "Invalid avatar URL")
+
+    result = await db.execute(select(User).where(User.id == current_user.id))
+    user = result.scalar_one()
+    user.avatar_url = body.avatar_url
+    await db.commit()
+
+    return await _build_profile(current_user.id, current_user.email, db)
+
+
+# ── Account deletion ─────────────────────────────────────────────────
+
+
+@router.delete("/me", status_code=status.HTTP_200_OK)
+async def delete_account(
+    current_user: UserProfile = Depends(get_current_user),
+    db: AsyncSession = Depends(get_session),
+) -> dict[str, bool]:
+    """Permanently delete the authenticated user's account.
+
+    Cascades: refresh tokens, OAuth accounts, subscription, and all memory
+    rows are deleted via SQLAlchemy relationship cascades.  Stripe subscription
+    is cancelled if active.
+    """
+    # Cancel Stripe subscription if present.
+    try:
+        from app.billing.stripe_service import stripe_service  # noqa: PLC0415
+        await stripe_service.cancel_subscription(current_user.id, db)
+    except HTTPException:
+        pass  # No subscription — that's fine
+
+    # Delete all memory rows (core, associative, episodic, proactive).
+    try:
+        from app.models import (  # noqa: PLC0415
+            MemoryAssociative, MemoryCore, MemoryEpisodic, MemoryProactive,
+        )
+        for model in (MemoryCore, MemoryAssociative, MemoryEpisodic, MemoryProactive):
+            await db.execute(
+                model.__table__.delete().where(model.user_id == current_user.id)
+            )
+    except Exception:
+        pass  # Non-critical — cascade on User will handle most
+
+    # Delete the user row — cascades handle refresh_tokens, oauth_accounts, subscription.
+    result = await db.execute(select(User).where(User.id == current_user.id))
+    user = result.scalar_one()
+    await db.delete(user)
+    await db.commit()
+
+    return {"ok": True}

app/api/routes/backup.py

@@ -1,171 +0,0 @@
-"""Backup routes: upload, download, history, and delete E2E-encrypted backups.
-
-Blobs are stored in S3 via BlobStore. Backup metadata is persisted in the
-PostgreSQL ``backup_metadata`` table.
-
-IMPORTANT: GET /history must be declared BEFORE GET / to avoid FastAPI
-treating "history" as a ``{backup_id}`` path parameter.
-"""
-
-from __future__ import annotations
-
-import uuid
-from email.utils import parsedate_to_datetime
-
-from fastapi import APIRouter, Depends, Header, HTTPException, Request, Response, status
-from sqlalchemy import func, select
-from sqlalchemy.ext.asyncio import AsyncSession
-
-from app.api.deps import get_current_user
-from app.billing.tier_manager import tier_manager
-from app.db import get_session
-from app.models import BackupMetadata as BackupMetadataModel
-from app.schemas import BackupMetadata, UserProfile
-from app.storage.blob_store import BlobStore
-from app.storage.encryption import reject_if_tampered
-
-router = APIRouter(prefix="/backup", tags=["backup"])
-
-_blob_store = BlobStore()
-
-
-async def _current_backup_bytes(user_id: str, db: AsyncSession) -> int:
-    """Return total backup bytes stored by *user_id*."""
-    result = await db.execute(
-        select(func.coalesce(func.sum(BackupMetadataModel.size_bytes), 0)).where(
-            BackupMetadataModel.user_id == user_id
-        )
-    )
-    return int(result.scalar_one())
-
-
-async def _check_backup_quota(
-    user: UserProfile, size_bytes: int, db: AsyncSession
-) -> None:
-    """Raise HTTP 402 if the upload would exceed the tier's backup limit."""
-    current = await _current_backup_bytes(user.id, db)
-    tier_manager.enforce_backup_quota(
-        user.tier, current_bytes=current, additional_bytes=size_bytes
-    )
-
-
-@router.put("")
-async def upload_backup(
-    request: Request,
-    x_backup_version: int = Header(..., alias="X-Backup-Version"),
-    x_backup_timestamp: int = Header(..., alias="X-Backup-Timestamp"),
-    x_backup_checksum: str = Header(..., alias="X-Backup-Checksum"),
-    current_user: UserProfile = Depends(get_current_user),
-    db: AsyncSession = Depends(get_session),
-) -> dict[str, bool]:
-    """Upload an E2E-encrypted backup blob.
-
-    Metadata is passed via custom headers; the raw body is the encrypted blob.
-    """
-    blob = await request.body()
-    reject_if_tampered(blob, x_backup_checksum)
-    await _check_backup_quota(current_user, len(blob), db)
-
-    s3_key = await _blob_store.upload(
-        current_user.id, "backup", str(x_backup_timestamp), blob, x_backup_checksum
-    )
-
-    row = BackupMetadataModel(
-        id=str(uuid.uuid4()),
-        user_id=current_user.id,
-        s3_key=s3_key,
-        version=x_backup_version,
-        timestamp=x_backup_timestamp,
-        checksum=x_backup_checksum,
-        size_bytes=len(blob),
-    )
-    db.add(row)
-    await db.commit()
-
-    return {"ok": True}
-
-
-@router.get("/history", response_model=list[BackupMetadata])
-async def backup_history(
-    current_user: UserProfile = Depends(get_current_user),
-    db: AsyncSession = Depends(get_session),
-) -> list[BackupMetadata]:
-    """Return backup metadata records for the authenticated user (no blob bytes)."""
-    result = await db.execute(
-        select(BackupMetadataModel)
-        .where(BackupMetadataModel.user_id == current_user.id)
-        .order_by(BackupMetadataModel.timestamp.desc())
-    )
-    rows = result.scalars().all()
-    return [
-        BackupMetadata(
-            version=r.version,
-            timestamp=r.timestamp,
-            checksum=r.checksum,
-            chunk_count=1,
-        )
-        for r in rows
-    ]
-
-
-@router.get("")
-async def download_backup(
-    request: Request,
-    current_user: UserProfile = Depends(get_current_user),
-    db: AsyncSession = Depends(get_session),
-) -> Response:
-    """Download the latest backup blob. Supports ``If-Modified-Since``."""
-    result = await db.execute(
-        select(BackupMetadataModel)
-        .where(BackupMetadataModel.user_id == current_user.id)
-        .order_by(BackupMetadataModel.timestamp.desc())
-        .limit(1)
-    )
-    latest = result.scalar_one_or_none()
-    if latest is None:
-        raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail="No backup found")
-
-    ims_header = request.headers.get("If-Modified-Since")
-    if ims_header:
-        try:
-            ims_dt = parsedate_to_datetime(ims_header)
-            ims_ms = int(ims_dt.timestamp() * 1000)
-            if latest.timestamp <= ims_ms:
-                return Response(status_code=status.HTTP_304_NOT_MODIFIED)
-        except Exception:
-            pass  # malformed header — ignore and serve the blob
-
-    blob = await _blob_store.download(current_user.id, latest.s3_key)
-    return Response(
-        content=blob,
-        media_type="application/octet-stream",
-        headers={
-            "X-Backup-Version": str(latest.version),
-            "X-Backup-Timestamp": str(latest.timestamp),
-            "X-Checksum": latest.checksum,
-        },
-    )
-
-
-@router.delete("/{backup_id}", response_model=dict)
-async def delete_backup(
-    backup_id: str,
-    current_user: UserProfile = Depends(get_current_user),
-    db: AsyncSession = Depends(get_session),
-) -> dict[str, bool]:
-    """Delete a specific backup by ID."""
-    result = await db.execute(
-        select(BackupMetadataModel).where(
-            BackupMetadataModel.id == backup_id,
-            BackupMetadataModel.user_id == current_user.id,
-        )
-    )
-    target = result.scalar_one_or_none()
-    if target is None:
-        raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail="Backup not found")
-
-    await _blob_store.delete(current_user.id, target.s3_key)
-    await db.delete(target)
-    await db.commit()
-
-    return {"ok": True}

app/api/routes/billing.py

@@ -83,3 +83,16 @@ async def cancel_subscription(
     """Cancel the active subscription."""
     await stripe_service.cancel_subscription(current_user.id, db)
     return {"ok": True}
+
+
+@router.get("/invoices", response_model=list[dict])
+async def list_invoices(
+    current_user: UserProfile = Depends(get_current_user),
+    db: AsyncSession = Depends(get_session),
+) -> list[dict[str, Any]]:
+    """Return billing history (invoices) from Stripe.
+
+    Returns an empty list when Stripe is not configured.
+    """
+    invoices = await stripe_service.list_invoices(current_user.id, db)
+    return invoices

app/api/routes/chat.py

@@ -1,4 +1,4 @@
-"""Chat routes: POST /chat (REST fallback).
+"""Chat routes: POST /chat (REST fallback) and POST /chat/embed (text → vector).
 
 WebSocket chat is handled by the unified device WS endpoint (/api/v1/ws/device).
 """
@@ -7,14 +7,30 @@ from __future__ import annotations
 
 from fastapi import APIRouter, Depends
 from fastapi.responses import JSONResponse
+from pydantic import BaseModel
 
 from app.api.deps import get_current_user
 from app.core.deep_agent import run_home
+from app.core.llm import embed
 from app.schemas import ChatRequest, UserProfile
 
 router = APIRouter(prefix="/chat", tags=["chat"])
 
 
+# ── Embed helpers ─────────────────────────────────────────────────────────
+
+
+class _EmbedRequest(BaseModel):
+    text: str
+
+
+class _EmbedResponse(BaseModel):
+    vector: list[float]
+
+
+# ── Endpoints ─────────────────────────────────────────────────────────────
+
+
 @router.post("")
 async def chat(
     body: ChatRequest,
@@ -27,3 +43,17 @@ async def chat(
         context=body.context.model_dump(),
     )
     return JSONResponse(content={"response": response})
+
+
+@router.post("/embed", response_model=_EmbedResponse)
+async def embed_text(
+    body: _EmbedRequest,
+    current_user: UserProfile = Depends(get_current_user),
+) -> _EmbedResponse:
+    """Generate a 1536-dim embedding vector for the given text.
+
+    Uses ``text-embedding-3-small`` via OpenAI.  Auth required (JWT).
+    Used by Electron (vectordb.ts) for local note search.
+    """
+    vector = await embed(body.text)
+    return _EmbedResponse(vector=vector)

app/api/routes/memory.py

@@ -0,0 +1,225 @@
+"""Memory management routes — view/edit/delete user memory tiers.
+
+All routes require authentication. Data is always user-scoped.
+"""
+
+from __future__ import annotations
+
+import logging
+from datetime import datetime, timezone
+from typing import Annotated
+
+from fastapi import APIRouter, Depends, Header, HTTPException, status
+from pydantic import BaseModel, Field
+from sqlalchemy import delete, select
+from sqlalchemy.ext.asyncio import AsyncSession
+
+from app.api.deps import get_current_user
+from app.core.memory_middleware import MemoryMiddleware
+from app.db import get_session
+from app.models import (
+    ExtractionQueue,
+    MemoryAssociative,
+    MemoryCore,
+    MemoryEpisodic,
+    MemoryProactive,
+    MemoryRelation,
+)
+from app.schemas import UserProfile
+
+router = APIRouter(prefix="/memory", tags=["memory"])
+
+logger = logging.getLogger(__name__)
+
+_ALLOWED_PREDICATES = {
+    "works_at",
+    "reports_to",
+    "stakeholder_of",
+    "last_contacted_on",
+    "owes_followup",
+    "manages",
+    "collaborates_with",
+    "owns",
+    "member_of",
+    "custom",
+}
+
+
+# ── Response schemas ─────────────────────────────────────────────────────────
+
+class RelationOut(BaseModel):
+    id: str
+    subject_label: str
+    subject_type: str
+    predicate: str
+    object_label: str
+    object_type: str
+    confidence: float
+    last_confirmed_at: int | None = None  # epoch ms
+
+
+class RelationPatch(BaseModel):
+    subject_label: str | None = None
+    object_label: str | None = None
+    predicate: str | None = None
+    confidence: float | None = Field(None, ge=0.0, le=1.0)
+
+
+class CoreAddBody(BaseModel):
+    key: str = Field(..., min_length=1, max_length=255)
+    value: str = Field(..., min_length=1)
+
+
+# ── Helpers ──────────────────────────────────────────────────────────────────
+
+def _relation_to_out(row: MemoryRelation) -> RelationOut:
+    last_ms: int | None = None
+    if row.last_confirmed_at is not None:
+        last_ms = int(row.last_confirmed_at.timestamp() * 1000)
+    return RelationOut(
+        id=row.id,
+        subject_label=row.subject_label,
+        subject_type=row.subject_type,
+        predicate=row.predicate,
+        object_label=row.object_label,
+        object_type=row.object_type,
+        confidence=row.confidence,
+        last_confirmed_at=last_ms,
+    )
+
+
+# ── Routes ───────────────────────────────────────────────────────────────────
+
+@router.get("/core", response_model=dict[str, str])
+async def get_core_memory(
+    current_user: UserProfile = Depends(get_current_user),
+    db: AsyncSession = Depends(get_session),
+) -> dict[str, str]:
+    """Return all core memory k/v pairs (plaintext) for the current user."""
+    mw = MemoryMiddleware(db)
+    blocks = await mw.list_core_blocks(current_user.id)
+    return {b["label"]: b["value"] for b in blocks}
+
+
+@router.delete("/core/{key}", status_code=status.HTTP_204_NO_CONTENT)
+async def delete_core_key(
+    key: str,
+    current_user: UserProfile = Depends(get_current_user),
+    db: AsyncSession = Depends(get_session),
+) -> None:
+    """Delete a single core memory key (GDPR Art. 17)."""
+    mw = MemoryMiddleware(db)
+    deleted = await mw.delete_core(current_user.id, key)
+    if not deleted:
+        raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail="Key not found")
+
+
+@router.post("/core", status_code=status.HTTP_201_CREATED, response_model=dict[str, str])
+async def add_core_key(
+    body: CoreAddBody,
+    current_user: UserProfile = Depends(get_current_user),
+    db: AsyncSession = Depends(get_session),
+) -> dict[str, str]:
+    """Add or overwrite a core memory key/value pair."""
+    mw = MemoryMiddleware(db)
+    await mw.update_core(current_user.id, body.key, body.value)
+    return {body.key: body.value}
+
+
+@router.get("/relational", response_model=list[RelationOut])
+async def get_relational_memory(
+    current_user: UserProfile = Depends(get_current_user),
+    db: AsyncSession = Depends(get_session),
+) -> list[RelationOut]:
+    """Return all relational memory rows for the current user."""
+    mw = MemoryMiddleware(db)
+    rows = await mw.query_relations(current_user.id, limit=200)
+    return [_relation_to_out(r) for r in rows]
+
+
+@router.patch("/relational/{relation_id}", response_model=RelationOut)
+async def patch_relation(
+    relation_id: str,
+    body: RelationPatch,
+    current_user: UserProfile = Depends(get_current_user),
+    db: AsyncSession = Depends(get_session),
+) -> RelationOut:
+    """Edit a relation row's labels, predicate, or confidence."""
+    if body.predicate is not None and body.predicate not in _ALLOWED_PREDICATES:
+        raise HTTPException(
+            status_code=status.HTTP_422_UNPROCESSABLE_ENTITY,
+            detail=f"predicate must be one of: {sorted(_ALLOWED_PREDICATES)}",
+        )
+
+    result = await db.execute(
+        select(MemoryRelation).where(
+            MemoryRelation.id == relation_id,
+            MemoryRelation.user_id == current_user.id,
+        )
+    )
+    row = result.scalar_one_or_none()
+    if row is None:
+        raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail="Relation not found")
+
+    if body.subject_label is not None:
+        row.subject_label = body.subject_label
+    if body.object_label is not None:
+        row.object_label = body.object_label
+    if body.predicate is not None:
+        row.predicate = body.predicate
+    if body.confidence is not None:
+        row.confidence = body.confidence
+        row.last_confirmed_at = datetime.now(timezone.utc)
+
+    await db.commit()
+    await db.refresh(row)
+    logger.info("memory: patch_relation user=%s relation=%s", current_user.id, relation_id)
+    return _relation_to_out(row)
+
+
+@router.delete("/relational/{relation_id}", status_code=status.HTTP_204_NO_CONTENT)
+async def delete_relation(
+    relation_id: str,
+    current_user: UserProfile = Depends(get_current_user),
+    db: AsyncSession = Depends(get_session),
+) -> None:
+    """Hard-delete a relation row (GDPR Art. 17)."""
+    result = await db.execute(
+        select(MemoryRelation).where(
+            MemoryRelation.id == relation_id,
+            MemoryRelation.user_id == current_user.id,
+        )
+    )
+    row = result.scalar_one_or_none()
+    if row is None:
+        raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail="Relation not found")
+    await db.delete(row)
+    await db.commit()
+    logger.info("memory: delete_relation user=%s relation=%s", current_user.id, relation_id)
+
+
+@router.post("/forget-all", status_code=status.HTTP_204_NO_CONTENT)
+async def forget_all(
+    x_confirm: Annotated[str | None, Header(alias="X-Confirm")] = None,
+    current_user: UserProfile = Depends(get_current_user),
+    db: AsyncSession = Depends(get_session),
+) -> None:
+    """Wipe all memory tiers for the current user (GDPR Art. 17).
+
+    Requires ``X-Confirm: true`` header. Does NOT delete the user account.
+    """
+    if x_confirm != "true":
+        raise HTTPException(
+            status_code=status.HTTP_400_BAD_REQUEST,
+            detail="Missing or invalid X-Confirm header. Send X-Confirm: true to confirm.",
+        )
+
+    uid = current_user.id
+    await db.execute(delete(MemoryCore).where(MemoryCore.user_id == uid))
+    await db.execute(delete(MemoryAssociative).where(MemoryAssociative.user_id == uid))
+    await db.execute(delete(MemoryEpisodic).where(MemoryEpisodic.user_id == uid))
+    await db.execute(delete(MemoryProactive).where(MemoryProactive.user_id == uid))
+    await db.execute(delete(MemoryRelation).where(MemoryRelation.user_id == uid))
+    await db.execute(delete(ExtractionQueue).where(ExtractionQueue.user_id == uid))
+    await db.commit()
+    logger.warning("memory: forget_all GDPR wipe user=%s", uid)

app/api/routes/plugins.py

@@ -1,148 +0,0 @@
-"""Plugins routes: browse and install plugins from the marketplace.
-
-Backed by ``PluginRegistry`` and ``RevenueShare`` service classes that
-persist data in the PostgreSQL ``plugins`` and ``revenue_events`` tables.
-"""
-
-from __future__ import annotations
-
-from typing import Any, Literal
-
-from fastapi import APIRouter, Depends, HTTPException, Query, status
-from pydantic import BaseModel
-from sqlalchemy import select
-from sqlalchemy.ext.asyncio import AsyncSession
-
-from app.api.deps import get_current_user
-from app.db import get_session
-from app.marketplace.plugin_registry import registry
-from app.marketplace.revenue_share import revenue_share
-from app.models import PluginInstallation, PluginReview as PluginReviewModel
-from app.schemas import PluginInstallRequest, PluginListResponse, PluginManifest, UserProfile
-
-router = APIRouter(prefix="/plugins", tags=["plugins"])
-
-
-# ── Tier gate ─────────────────────────────────────────────────────────
-
-def _require_plugin_tier(user: UserProfile) -> None:
-    """Raise HTTP 403 for users below Power tier."""
-    if user.tier not in ("power", "team"):
-        raise HTTPException(
-            status_code=status.HTTP_403_FORBIDDEN,
-            detail="Plugin marketplace requires Power tier or above",
-        )
-
-
-# ── Local detail schema ────────────────────────────────────────────────
-
-class _PluginDetail(BaseModel):
-    plugin: PluginManifest
-    install_count: int
-    ratings: list[Any]
-
-
-# ── Routes ────────────────────────────────────────────────────────────
-
-@router.get("", response_model=PluginListResponse)
-async def list_plugins(
-    category: str | None = Query(default=None),
-    q: str | None = Query(default=None),
-    page: int = Query(default=1, ge=1),
-    sort: Literal["rating", "installs", "newest"] = Query(default="newest"),
-    current_user: UserProfile = Depends(get_current_user),
-    db: AsyncSession = Depends(get_session),
-) -> PluginListResponse:
-    """Browse the plugin marketplace. Requires Power tier or above."""
-    _require_plugin_tier(current_user)
-    return await registry.list_plugins(db, category=category, query=q, page=page, sort=sort)
-
-
-@router.get("/{plugin_id}", response_model=_PluginDetail)
-async def get_plugin(
-    plugin_id: str,
-    current_user: UserProfile = Depends(get_current_user),
-    db: AsyncSession = Depends(get_session),
-) -> _PluginDetail:
-    """Get full plugin details including install count. Requires Power tier or above."""
-    _require_plugin_tier(current_user)
-    entry = await registry.get_plugin(db, plugin_id)
-    if entry is None:
-        raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail="Plugin not found")
-
-    # Fetch review ratings for this plugin
-    review_result = await db.execute(
-        select(PluginReviewModel).where(PluginReviewModel.plugin_id == plugin_id)
-    )
-    reviews = review_result.scalars().all()
-    ratings = [
-        {
-            "reviewer_id": r.reviewer_id,
-            "decision": r.decision,
-            "notes": r.notes,
-            "reviewed_at": int(r.reviewed_at.timestamp() * 1000) if r.reviewed_at else None,
-        }
-        for r in reviews
-    ]
-
-    return _PluginDetail(
-        plugin=entry["manifest"],
-        install_count=entry["install_count"],
-        ratings=ratings,
-    )
-
-
-@router.post("/{plugin_id}/install", response_model=dict)
-async def install_plugin(
-    plugin_id: str,
-    body: PluginInstallRequest,  # noqa: ARG001 — reserved for future fields
-    current_user: UserProfile = Depends(get_current_user),
-    db: AsyncSession = Depends(get_session),
-) -> dict[str, Any]:
-    """Install a plugin. Triggers Stripe Connect revenue split for paid plugins.
-
-    Requires Power tier or above.
-    """
-    _require_plugin_tier(current_user)
-    entry = await registry.get_plugin(db, plugin_id)
-    if entry is None:
-        raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail="Plugin not found")
-
-    # Record the installation in plugin_installations
-    installation = PluginInstallation(
-        plugin_id=plugin_id,
-        user_id=current_user.id,
-    )
-    db.add(installation)
-    await db.flush()
-
-    await revenue_share.record_install(
-        db,
-        plugin_id=plugin_id,
-        user_id=current_user.id,
-        amount_cents=entry["manifest"].price_cents,
-    )
-
-    download_url = f"https://cdn.adiuva.app/plugins/{plugin_id}/package.zip"
-    return {"ok": True, "download_url": download_url}
-
-
-@router.delete("/{plugin_id}/install", response_model=dict)
-async def uninstall_plugin(
-    plugin_id: str,
-    current_user: UserProfile = Depends(get_current_user),
-    db: AsyncSession = Depends(get_session),
-) -> dict[str, bool]:
-    """Unregister a plugin installation."""
-    result = await db.execute(
-        select(PluginInstallation).where(
-            PluginInstallation.plugin_id == plugin_id,
-            PluginInstallation.user_id == current_user.id,
-        )
-    )
-    installation = result.scalar_one_or_none()
-    if installation is not None:
-        await db.delete(installation)
-        await db.commit()
-    await registry.record_uninstall(db, plugin_id)
-    return {"ok": True}

app/api/routes/storage.py

@@ -1,195 +0,0 @@
-"""Storage routes: CRUD for E2E-encrypted cloud records.
-
-Blobs are stored in S3 via BlobStore. Record metadata is persisted in the
-PostgreSQL ``storage_records`` table.
-"""
-
-from __future__ import annotations
-
-import uuid
-
-from fastapi import APIRouter, Depends, HTTPException, Query, Response, status
-from pydantic import BaseModel
-from sqlalchemy import func, select
-from sqlalchemy.ext.asyncio import AsyncSession
-
-from app.api.deps import get_current_user
-from app.billing.tier_manager import tier_manager
-from app.db import get_session
-from app.models import StorageRecord
-from app.schemas import StorageRecordCreate, StorageRecordUpdate, UserProfile
-from app.storage.blob_store import BlobStore
-from app.storage.encryption import reject_if_tampered
-
-router = APIRouter(prefix="/storage", tags=["storage"])
-
-_blob_store = BlobStore()
-
-
-# ── Local response schemas ─────────────────────────────────────────────
-
-class _CreateResponse(BaseModel):
-    id: str
-    created_at: int
-
-
-class _RecordMeta(BaseModel):
-    id: str
-    table: str
-    checksum: str
-    created_at: int
-    updated_at: int
-
-
-# ── Helpers ────────────────────────────────────────────────────────────
-
-async def _current_usage_bytes(user_id: str, db: AsyncSession) -> int:
-    """Return total bytes stored by *user_id*."""
-    result = await db.execute(
-        select(func.coalesce(func.sum(StorageRecord.size_bytes), 0)).where(
-            StorageRecord.user_id == user_id
-        )
-    )
-    return int(result.scalar_one())
-
-
-async def _check_quota(user: UserProfile, additional_bytes: int, db: AsyncSession) -> None:
-    """Raise HTTP 402 if adding *additional_bytes* would exceed the tier limit."""
-    current = await _current_usage_bytes(user.id, db)
-    tier_manager.enforce_quota(user.tier, current_bytes=current, additional_bytes=additional_bytes)
-
-
-async def _get_record_for_user(
-    record_id: str, user_id: str, db: AsyncSession
-) -> StorageRecord:
-    """Look up a record and verify ownership. Returns 404 on mismatch
-    to prevent user enumeration attacks."""
-    result = await db.execute(
-        select(StorageRecord).where(
-            StorageRecord.id == record_id, StorageRecord.user_id == user_id
-        )
-    )
-    record = result.scalar_one_or_none()
-    if record is None:
-        raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail="Record not found")
-    return record
-
-
-# ── Routes ─────────────────────────────────────────────────────────────
-
-@router.post("/records", response_model=_CreateResponse, status_code=status.HTTP_201_CREATED)
-async def create_record(
-    body: StorageRecordCreate,
-    current_user: UserProfile = Depends(get_current_user),
-    db: AsyncSession = Depends(get_session),
-) -> _CreateResponse:
-    """Upload a new E2E-encrypted blob. Verifies checksum before storing."""
-    reject_if_tampered(body.blob, body.checksum)
-    await _check_quota(current_user, len(body.blob), db)
-
-    record_id = str(uuid.uuid4())
-
-    s3_key = await _blob_store.upload(
-        current_user.id, body.table, record_id, body.blob, body.checksum
-    )
-
-    record = StorageRecord(
-        id=record_id,
-        user_id=current_user.id,
-        table_name=body.table,
-        s3_key=s3_key,
-        checksum=body.checksum,
-        size_bytes=len(body.blob),
-    )
-    db.add(record)
-    await db.commit()
-    await db.refresh(record)
-
-    created_at_ms = int(record.created_at.timestamp() * 1000)
-    return _CreateResponse(id=record_id, created_at=created_at_ms)
-
-
-@router.get("/records", response_model=list[_RecordMeta])
-async def list_records(
-    table: str | None = Query(default=None),
-    page: int = Query(default=1, ge=1),
-    limit: int = Query(default=50, ge=1, le=200),
-    current_user: UserProfile = Depends(get_current_user),
-    db: AsyncSession = Depends(get_session),
-) -> list[_RecordMeta]:
-    """List record metadata for the authenticated user. Blob bytes are never returned."""
-    query = select(StorageRecord).where(StorageRecord.user_id == current_user.id)
-    if table is not None:
-        query = query.where(StorageRecord.table_name == table)
-    query = query.offset((page - 1) * limit).limit(limit)
-
-    result = await db.execute(query)
-    rows = result.scalars().all()
-
-    return [
-        _RecordMeta(
-            id=r.id,
-            table=r.table_name,
-            checksum=r.checksum,
-            created_at=int(r.created_at.timestamp() * 1000),
-            updated_at=int(r.updated_at.timestamp() * 1000),
-        )
-        for r in rows
-    ]
-
-
-@router.get("/records/{record_id}")
-async def download_record(
-    record_id: str,
-    current_user: UserProfile = Depends(get_current_user),
-    db: AsyncSession = Depends(get_session),
-) -> Response:
-    """Download an E2E-encrypted blob. Returns raw bytes with ``X-Checksum`` header."""
-    record = await _get_record_for_user(record_id, current_user.id, db)
-    blob = await _blob_store.download(current_user.id, record.s3_key)
-    return Response(
-        content=blob,
-        media_type="application/octet-stream",
-        headers={"X-Checksum": record.checksum},
-    )
-
-
-@router.put("/records/{record_id}", response_model=dict)
-async def update_record(
-    record_id: str,
-    body: StorageRecordUpdate,
-    current_user: UserProfile = Depends(get_current_user),
-    db: AsyncSession = Depends(get_session),
-) -> dict[str, bool]:
-    """Replace the blob for an existing record. Verifies checksum before storing."""
-    record = await _get_record_for_user(record_id, current_user.id, db)
-    reject_if_tampered(body.blob, body.checksum)
-
-    delta = len(body.blob) - record.size_bytes
-    if delta > 0:
-        await _check_quota(current_user, delta, db)
-
-    s3_key = await _blob_store.upload(
-        current_user.id, record.table_name, record_id, body.blob, body.checksum
-    )
-
-    record.s3_key = s3_key
-    record.checksum = body.checksum
-    record.size_bytes = len(body.blob)
-    await db.commit()
-
-    return {"ok": True}
-
-
-@router.delete("/records/{record_id}", response_model=dict)
-async def delete_record(
-    record_id: str,
-    current_user: UserProfile = Depends(get_current_user),
-    db: AsyncSession = Depends(get_session),
-) -> dict[str, bool]:
-    """Delete a record and its S3 blob."""
-    record = await _get_record_for_user(record_id, current_user.id, db)
-    await _blob_store.delete(current_user.id, record.s3_key)
-    await db.delete(record)
-    await db.commit()
-    return {"ok": True}

app/api/routes/vectors.py

@@ -1,79 +0,0 @@
-"""Vectors routes: upsert, search, delete cloud vector store entries, and embed text."""
-
-from __future__ import annotations
-
-from fastapi import APIRouter, Depends
-from pydantic import BaseModel
-
-from app.api.deps import get_current_user
-from app.core.llm import embed
-from app.schemas import (
-    UserProfile,
-    VectorSearchRequest,
-    VectorSearchResponse,
-    VectorUpsertRequest,
-)
-from app.storage.encryption import reject_if_tampered
-from app.storage.vector_store import VectorStore
-
-router = APIRouter(prefix="/storage", tags=["vectors"])
-
-_vector_store = VectorStore()
-
-
-class _VectorDeleteRequest(BaseModel):
-    ids: list[str]
-
-
-class _EmbedRequest(BaseModel):
-    text: str
-
-
-class _EmbedResponse(BaseModel):
-    vector: list[float]
-
-
-@router.post("/vectors/upsert", response_model=dict)
-async def upsert_vectors(
-    body: VectorUpsertRequest,
-    current_user: UserProfile = Depends(get_current_user),
-) -> dict[str, int]:
-    """Verify checksums and store encrypted vectors in the user-scoped namespace."""
-    for item in body.vectors:
-        reject_if_tampered(item.blob, item.checksum)
-    await _vector_store.upsert(current_user.id, body.vectors)
-    return {"upserted": len(body.vectors)}
-
-
-@router.post("/vectors/search", response_model=VectorSearchResponse)
-async def search_vectors(
-    body: VectorSearchRequest,
-    current_user: UserProfile = Depends(get_current_user),
-) -> VectorSearchResponse:
-    """Search the user-scoped vector namespace with an encrypted query blob."""
-    results = await _vector_store.search(current_user.id, body.query_blob, body.top_k)
-    return VectorSearchResponse(results=results)
-
-
-@router.delete("/vectors", response_model=dict)
-async def delete_vectors(
-    body: _VectorDeleteRequest,
-    current_user: UserProfile = Depends(get_current_user),
-) -> dict[str, bool]:
-    """Delete vectors by ID, scoped to the authenticated user."""
-    await _vector_store.delete(current_user.id, body.ids)
-    return {"ok": True}
-
-
-@router.post("/vectors/embed", response_model=_EmbedResponse)
-async def embed_text(
-    body: _EmbedRequest,
-    current_user: UserProfile = Depends(get_current_user),
-) -> _EmbedResponse:
-    """Generate a 1536-dim embedding vector for the given text.
-
-    Uses ``text-embedding-3-small`` via OpenAI.  Auth required (JWT).
-    Used by backend tools (note_agent) and Electron (vectordb.ts) alike.
-    """
-    vector = await embed(body.text)
-    return _EmbedResponse(vector=vector)

app/auth/__init__.py

@@ -0,0 +1 @@
+"OAuth provider abstractions and utilities."

app/auth/oauth_providers.py

@@ -0,0 +1,135 @@
+"""OAuth 2.0 + PKCE provider abstractions.
+
+Each provider implements a three-step flow designed for a desktop (public) client:
+
+  1. get_authorization_url(state, code_challenge) → str
+       Build the provider's consent-screen URL.  State and code_challenge are
+       generated server-side; the client opens this URL in the system browser.
+
+  2. exchange_code(code, code_verifier, redirect_uri) → dict
+       Exchange the short-lived authorization code for an access token.
+       The code_verifier proves ownership of the PKCE challenge.
+
+  3. get_userinfo(access_token) → OAuthUserInfo
+       Fetch the canonical user identity from the provider.
+
+Currently supported providers:
+  - GoogleOAuthProvider  (scope: openid email profile)
+
+Adding a new provider:
+  - Implement the three methods above.
+  - Register in _PROVIDERS inside routes/auth.py.
+"""
+
+from __future__ import annotations
+
+import base64
+import hashlib
+import os
+import urllib.parse
+from dataclasses import dataclass
+
+import httpx
+
+
+# ── Data transfer objects ─────────────────────────────────────────────
+
+
+@dataclass
+class OAuthUserInfo:
+    """Normalized user identity returned by any provider."""
+
+    provider_user_id: str
+    email: str
+    email_verified: bool
+    avatar_url: str | None
+    name: str | None
+
+
+# ── PKCE helpers ──────────────────────────────────────────────────────
+
+
+def generate_pkce_pair() -> tuple[str, str]:
+    """Generate a (code_verifier, code_challenge) pair for PKCE S256.
+
+    The code_verifier is a random 32-byte URL-safe base64 string.
+    The code_challenge is SHA-256(code_verifier) base64url-encoded (no padding).
+    """
+    code_verifier = base64.urlsafe_b64encode(os.urandom(32)).rstrip(b"=").decode()
+    digest = hashlib.sha256(code_verifier.encode()).digest()
+    code_challenge = base64.urlsafe_b64encode(digest).rstrip(b"=").decode()
+    return code_verifier, code_challenge
+
+
+# ── Google provider ───────────────────────────────────────────────────
+
+
+class GoogleOAuthProvider:
+    """Google OAuth 2.0 provider (openid email profile scope).
+
+    Uses Google's standard authorization endpoint with PKCE S256.
+    Does NOT use google-auth-oauthlib to keep the flow generic and async.
+    """
+
+    name = "google"
+
+    _AUTH_URL = "https://accounts.google.com/o/oauth2/v2/auth"
+    _TOKEN_URL = "https://oauth2.googleapis.com/token"
+    _USERINFO_URL = "https://www.googleapis.com/oauth2/v3/userinfo"
+
+    def __init__(self, client_id: str, client_secret: str, redirect_uri: str) -> None:
+        self.client_id = client_id
+        self.client_secret = client_secret
+        self.redirect_uri = redirect_uri
+
+    def get_authorization_url(self, state: str, code_challenge: str) -> str:
+        """Build the Google consent-screen URL."""
+        params = {
+            "client_id": self.client_id,
+            "redirect_uri": self.redirect_uri,
+            "response_type": "code",
+            "scope": "openid email profile",
+            "state": state,
+            "code_challenge": code_challenge,
+            "code_challenge_method": "S256",
+            "access_type": "offline",
+            "prompt": "select_account",
+        }
+        return f"{self._AUTH_URL}?{urllib.parse.urlencode(params)}"
+
+    async def exchange_code(
+        self, code: str, code_verifier: str, redirect_uri: str
+    ) -> dict:
+        """Exchange authorization code for an access token."""
+        async with httpx.AsyncClient() as client:
+            response = await client.post(
+                self._TOKEN_URL,
+                data={
+                    "client_id": self.client_id,
+                    "client_secret": self.client_secret,
+                    "code": code,
+                    "code_verifier": code_verifier,
+                    "grant_type": "authorization_code",
+                    "redirect_uri": redirect_uri,
+                },
+            )
+        response.raise_for_status()
+        return response.json()
+
+    async def get_userinfo(self, access_token: str) -> OAuthUserInfo:
+        """Fetch the authenticated user's identity from Google."""
+        async with httpx.AsyncClient() as client:
+            response = await client.get(
+                self._USERINFO_URL,
+                headers={"Authorization": f"Bearer {access_token}"},
+            )
+        response.raise_for_status()
+        data = response.json()
+
+        return OAuthUserInfo(
+            provider_user_id=data["sub"],
+            email=data["email"],
+            email_verified=data.get("email_verified", False),
+            avatar_url=data.get("picture"),
+            name=data.get("name"),
+        )

app/billing/stripe_service.py

@@ -43,8 +43,8 @@ class StripeService:
         self,
         user_id: str,
         tier: str,
-        success_url: str = "https://app.adiuva.app/billing/success?session_id={CHECKOUT_SESSION_ID}",
+        success_url: str = "https://app.adiuvai.app/billing/success?session_id={CHECKOUT_SESSION_ID}",
-        cancel_url: str = "https://app.adiuva.app/billing/cancel",
+        cancel_url: str = "https://app.adiuvai.app/billing/cancel",
     ) -> str:
         """Create a Stripe checkout session and return the URL.
 
@@ -200,6 +200,45 @@ class StripeService:
         sub.status = "canceled"
         await db.commit()
 
+    async def list_invoices(
+        self, user_id: str, db: AsyncSession, limit: int = 24
+    ) -> list[dict[str, Any]]:
+        """Return recent invoices for the user from Stripe.
+
+        Returns an empty list when Stripe is not configured or the user has
+        no ``stripe_customer_id``.
+        """
+        if not self._configured():
+            return []
+
+        from app.models import User  # noqa: PLC0415
+
+        result = await db.execute(
+            select(User.stripe_customer_id).where(User.id == user_id)
+        )
+        customer_id = result.scalar_one_or_none()
+        if not customer_id:
+            return []
+
+        try:
+            s = self._client()
+            invoices = s.Invoice.list(customer=customer_id, limit=limit)
+            return [
+                {
+                    "id": inv.id,
+                    "amount_due": inv.amount_due,
+                    "amount_paid": inv.amount_paid,
+                    "currency": inv.currency,
+                    "status": inv.status,
+                    "created": inv.created * 1000,  # epoch ms
+                    "invoice_url": inv.hosted_invoice_url,
+                    "invoice_pdf": inv.invoice_pdf,
+                }
+                for inv in invoices.auto_paging_iter()
+            ]
+        except Exception:
+            return []
+
     # ── Private DB helpers ───────────────────────────────────────────────
 
     async def _upsert_subscription(

app/billing/tier_manager.py

@@ -22,45 +22,49 @@ FEATURES: dict[str, dict[str, Any]] = {
         "agents": 3,
         "batch_active": 2,
         "batch_runs_per_day": 5,
-        "cloud_storage_gb": 0,
-        "backup_gb": 0,
         "providers": 1,
         "batch_builder": False,
-        "plugin_marketplace": False,
         "sso": False,
+        "real_embeddings": False,       # keyword fallback only
+        "realtime_extraction": False,   # batch queue (Phase 2)
+        "relational_memory": False,     # relational tier (Phase 3) — Pro+
+        "proactive_mining": False,      # Power+ only (Phase 5)
     },
     "pro": {
         "agents": -1,           # unlimited
         "batch_active": 10,
         "batch_runs_per_day": 50,
-        "cloud_storage_gb": 5,
-        "backup_gb": 5,
         "providers": -1,
         "batch_builder": False,
-        "plugin_marketplace": False,
         "sso": False,
+        "real_embeddings": True,        # pgvector cosine search
+        "realtime_extraction": True,    # fire-and-forget asyncio.create_task
+        "relational_memory": True,      # person/project predicates
+        "proactive_mining": False,      # Power+ only (Phase 5)
     },
     "power": {
         "agents": -1,
         "batch_active": -1,     # unlimited
         "batch_runs_per_day": -1,  # unlimited
-        "cloud_storage_gb": 25,
-        "backup_gb": 25,
         "providers": -1,
         "batch_builder": True,
-        "plugin_marketplace": True,
         "sso": False,
+        "real_embeddings": True,
+        "realtime_extraction": True,
+        "relational_memory": True,      # all predicates incl. custom
+        "proactive_mining": True,       # scheduled pattern mining (Phase 5)
     },
     "team": {
         "agents": -1,
         "batch_active": -1,
         "batch_runs_per_day": -1,  # unlimited
-        "cloud_storage_gb": -1,  # unlimited
-        "backup_gb": -1,         # unlimited
         "providers": -1,
         "batch_builder": True,
-        "plugin_marketplace": True,
         "sso": True,
+        "real_embeddings": True,
+        "realtime_extraction": True,
+        "relational_memory": True,      # all predicates incl. custom
+        "proactive_mining": True,       # scheduled pattern mining (Phase 5)
     },
 }
 
@@ -125,71 +129,6 @@ class TierManager:
         """Return the requests-per-minute limit for ``tier``."""
         return RATE_LIMITS.get(tier, RATE_LIMITS["free"])
 
-    # ── Storage quota ────────────────────────────────────────────────────
-
-    def enforce_quota(
-        self,
-        tier: BillingTier,
-        current_bytes: int = 0,
-        additional_bytes: int = 0,
-    ) -> None:
-        """Raise ``HTTP 402`` if the user would exceed their cloud storage quota.
-
-        ``tier`` is the caller's current tier (from ``current_user.tier``).
-        ``current_bytes`` is the total bytes already stored (queried by caller).
-        """
-        limit_gb: int = FEATURES[tier]["cloud_storage_gb"]
-        if limit_gb == 0:
-            raise HTTPException(
-                status_code=status.HTTP_402_PAYMENT_REQUIRED,
-                detail=f"Cloud storage is not available on the '{tier}' tier",
-            )
-        if limit_gb == -1:
-            return  # unlimited
-        limit_bytes = limit_gb * 1024 ** 3
-        if current_bytes + additional_bytes > limit_bytes:
-            raise HTTPException(
-                status_code=status.HTTP_402_PAYMENT_REQUIRED,
-                detail=f"Storage quota exceeded for tier '{tier}'",
-            )
-
-    def enforce_backup_quota(
-        self,
-        tier: BillingTier,
-        current_bytes: int = 0,
-        additional_bytes: int = 0,
-    ) -> None:
-        """Raise ``HTTP 402`` if the user would exceed their backup quota."""
-        limit_gb: int = FEATURES[tier]["backup_gb"]
-        if limit_gb == 0:
-            raise HTTPException(
-                status_code=status.HTTP_402_PAYMENT_REQUIRED,
-                detail=f"Backup is not available on the '{tier}' tier",
-            )
-        if limit_gb == -1:
-            return  # unlimited
-        limit_bytes = limit_gb * 1024 ** 3
-        if current_bytes + additional_bytes > limit_bytes:
-            raise HTTPException(
-                status_code=status.HTTP_402_PAYMENT_REQUIRED,
-                detail=f"Backup quota exceeded for tier '{tier}'",
-            )
-
-    def check_quota(
-        self,
-        tier: BillingTier,
-        current_bytes: int = 0,
-        additional_bytes: int = 0,
-    ) -> bool:
-        """Return ``True`` if the user can store ``additional_bytes`` more data."""
-        limit_gb: int = FEATURES[tier]["cloud_storage_gb"]
-        if limit_gb == 0:
-            return False
-        if limit_gb == -1:
-            return True
-        limit_bytes = limit_gb * 1024 ** 3
-        return current_bytes + additional_bytes <= limit_bytes
-
 
 # Module-level singleton shared across the app.
 tier_manager = TierManager()

app/config/settings.py

@@ -3,7 +3,7 @@ from pydantic_settings import BaseSettings, SettingsConfigDict
 
 
 class Settings(BaseSettings):
-    DATABASE_URL: str = "postgresql+asyncpg://postgres:postgres@localhost:5432/adiuva"
+    DATABASE_URL: str = "postgresql+asyncpg://postgres:postgres@localhost:5432/adiuvai"
     JWT_SECRET: str = "change-me-in-production"
     JWT_ALGORITHM: str = "HS256"
     JWT_ACCESS_TOKEN_EXPIRE_MINUTES: int = 30
@@ -12,26 +12,25 @@ class Settings(BaseSettings):
     STRIPE_SECRET_KEY: str = ""
     STRIPE_WEBHOOK_SECRET: str = ""
 
-    S3_BUCKET: str = ""
-    S3_REGION: str = "us-east-1"
-    S3_ENDPOINT_URL: str = ""
-    AWS_ACCESS_KEY_ID: str = ""
-    AWS_SECRET_ACCESS_KEY: str = ""
-
-    PINECONE_API_KEY: str = ""
-    PINECONE_INDEX: str = "adiuva"
-    QDRANT_URL: str = ""
-    QDRANT_API_KEY: str = ""
-
     OPENAI_API_KEY: str = ""
     ANTHROPIC_API_KEY: str = ""
     GOOGLE_API_KEY: str = ""
     CEREBRAS_API_KEY: str = ""
 
     LLM_MODEL: str = "gpt-4o"
-    LLM_ROUTER_MODEL: str = "gpt-4o-mini"
     LLM_EMBED_MODEL: str = "text-embedding-3-small"
 
+    # Per-agent model overrides. Leave empty to fall back to LLM_MODEL.
+    LLM_MODEL_CLASSIFIER: str = ""        # _infer_floating_domain (intent routing)
+    LLM_MODEL_HOME_AGENT: str = ""        # home-agent (run_single_agent / stream)
+    LLM_MODEL_FLOATING_AGENT: str = ""    # floating-agent (contextual chat)
+    LLM_MODEL_UNIFIED_PROCESSOR: str = "" # unified-processor (agent_runner)
+    LLM_MODEL_CLOUD_PROCESSOR: str = ""   # cloud-processor (agent_runner)
+    LLM_MODEL_SETUP_AGENT: str = ""       # agent-setup journey
+    LLM_MODEL_MEMORY_EXTRACTOR: str = ""  # memory-extractor (Phase 2 extract/decide)
+    LLM_MODEL_MEMORY_MINER: str = ""      # memory-miner (Phase 5 proactive mining)
+    LLM_MODEL_MEMORY_AUDITOR: str = ""    # memory-auditor (Phase 7 weekly audit)
+
     # GitHub Copilot OAuth token storage directory.
     # Leave empty to use the LiteLLM default (~/.config/litellm/github_copilot).
     # In Docker, set this to a path backed by a named volume so tokens survive restarts.
@@ -45,16 +44,39 @@ class Settings(BaseSettings):
     # MS_TENANT_ID: set to 'common' to allow multi-tenant (personal + work accounts).
     MS_TENANT_ID: str = "common"
 
+    # Google Login OAuth credentials — scope: openid email profile.
+    # Separate from GMAIL_CLIENT_ID/SECRET (which uses gmail.readonly scope).
+    GOOGLE_AUTH_CLIENT_ID: str = ""
+    GOOGLE_AUTH_CLIENT_SECRET: str = ""
+    # The redirect URI registered in Google Cloud Console.
+    # Google redirects here after consent; this backend route then bounces to
+    # the adiuvai:// deep link so the Electron app receives the code.
+    # Dev:  http://localhost:8000/api/v1/auth/oauth/google/web-callback
+    # Prod: https://api.adiuvai.com/api/v1/auth/oauth/google/web-callback
+    OAUTH_REDIRECT_URI: str = "http://localhost:8000/api/v1/auth/oauth/google/web-callback"
+
     # Fernet key (URL-safe base64, 32-byte key) for at-rest encryption of OAuth
     # tokens stored in cloud_agent_configs.oauth_token_encrypted.
     # Generate with: from cryptography.fernet import Fernet; Fernet.generate_key()
     OAUTH_ENCRYPTION_KEY: str = ""
 
-    CORS_ORIGINS: list[str] = ["app://.", "http://localhost:3000", "http://localhost:5173"]
+    CORS_ORIGINS: list[str] = [
+        "app://.",
+        "http://localhost:3000",
+        "http://localhost:5173",
+        "http://localhost:4173",      # Vite preview (web SPA)
+        "https://app.adiuvai.com",    # Production web portal
+    ]
+
+    LANGFUSE_SECRET_KEY: str = ""
+    LANGFUSE_PUBLIC_KEY: str = ""
+    LANGFUSE_BASE_URL: str = "https://cloud.langfuse.com"
+
+    SCHEDULER_ENABLED: bool = True
 
     ENV: Literal["dev", "prod"] = "dev"
 
-    model_config = SettingsConfigDict(env_file=".env", env_file_encoding="utf-8")
+    model_config = SettingsConfigDict(env_file=".env", env_file_encoding="utf-8", extra="ignore")
 
 
 settings = Settings()

app/core/agent_runner.py

@@ -2,12 +2,12 @@
 
 Drives two agent types:
 
-* **Local directory agent** — two-step execution per file:
+* **Local directory agent** — V2 unified flow per file:
-  Step 1 (Classification) uses code to fetch all projects and asks the LLM
+  Phase A (Detect + Preprocess, zero LLM): Python detects the content type
-  to identify which project the file belongs to and which domains are relevant.
+  and strips markup/noise, producing clean text + metadata.
-  Step 2 (Processing) fetches existing entities for that project/domains via
+  Phase B (Single LLM call with tools): the LLM identifies the project,
-  code and runs an LLM with tools — existing data in context enforces
+  checks for duplicates via list_* tools, and creates/updates records.
-  update-first naturally.
+  ``items_created`` is counted from ``create_*`` tool calls.
 
 * **Cloud connector agent** — fetches data from third-party APIs (Gmail,
   Teams, Outlook) and pushes extracted items to Electron.
@@ -29,7 +29,7 @@ from __future__ import annotations
 import asyncio
 import json
 import logging
-import uuid
+import os
 from datetime import datetime, timedelta, timezone
 from typing import Any
 
@@ -43,7 +43,9 @@ from app.agents.project_agent import PROJECT_TOOLS
 from app.agents.task_agent import TASK_TOOLS
 from app.agents.timeline_agent import TIMELINE_TOOLS
 from app.core.device_manager import DeviceConnectionManager
-from app.core.llm import get_llm
+from app.core.langfuse_client import compile_prompt, extract_usage, get_langfuse, get_prompt_or_fallback, langfuse_context
+from app.core.llm import get_agent_llm, model_for_agent
+from app.core.preprocessors import detect_content_type, preprocess
 from app.core.ws_context import clear_client_executor, execute_on_client, set_client_executor
 from app.db import async_session
 from app.models import AgentRunLog, CloudAgentConfig, LocalAgentConfig
@@ -70,97 +72,52 @@ _MAX_PROCESSING_STEPS: int = 12
 _MAX_SCAN_DEPTH: int = 5
 
 # ── Data-type to tool mapping ─────────────────────────────────────────────
-# NOTE: "projects" is intentionally excluded — project creation/assignment is
-# handled in code by the runner, never delegated to the Step 2 LLM.
 
 _DATA_TYPE_TOOLS: dict[str, list[Any]] = {
     "tasks": TASK_TOOLS,
     "notes": NOTE_TOOLS,
     "timelines": TIMELINE_TOOLS,
+    "timelineEvents": TIMELINE_TOOLS,
+    "projects": PROJECT_TOOLS,
 }
 
-# ── Step 1: Classification prompt ─────────────────────────────────────────
+# ── V2: Unified processing prompt (hot-swappable via Langfuse "unified_processing") ──
 
-_DOMAIN_DESCRIPTIONS: dict[str, str] = {
+_UNIFIED_PROCESSING_PROMPT = """\
-    "tasks": (
-        "Action items, to-dos, deliverables — anything that describes work to be done, "
-        "assigned to someone, or tracked with a due date or status."
-    ),
-    "notes": (
-        "Documentation, meeting notes, summaries, reference material — "
-        "written content meant to be read and referenced rather than acted on."
-    ),
-    "timelines": (
-        "Project milestones, deadlines, scheduled events — "
-        "specific dates that mark a point in the progress of a project."
-    ),
-    "projects": (
-        "High-level project entities — only relevant if the file clearly introduces "
-        "a new project or updates the scope of an existing one."
-    ),
-}
-
-_STEP1_SYSTEM_PROMPT = """\
-You are a file classifier for a freelance project management tool.
-
-Your job is to match a file to an existing project and identify which data domains to extract.
-
-## Project matching rules (STRICT — follow in order)
-
-1. Search the file content for any mention of a project name, client name, acronym, or topic
-   that overlaps with the existing projects listed below.
-2. The match does NOT need to be exact — partial name, abbreviation, or topic similarity is enough.
-3. STRONGLY PREFER matching an existing project. Only return "new" as an absolute last resort
-   when the file has zero meaningful connection to any listed project.
-4. When in doubt, pick the closest match from the list.
-
-## Response format
-
-Respond ONLY with a JSON object — no markdown, no explanation:
-
-{{"project_id": "<exact id from the list below, or new>", "new_project_name": "<concise 2-5 word name, only when project_id is new>", "domains": ["tasks", "notes"]}}
-
-## Domain definitions (only consider domains in the allowed list)
-
-{domain_definitions}
-
-## Existing projects
-
-{projects_list}
-"""
-
-# ── Step 2: Processing prompt ─────────────────────────────────────────────
-
-_PROCESSING_SYSTEM_PROMPT = """\
 You are a data extraction assistant for a freelance project management tool.
 
-Your task: extract structured data from the file content and persist it using the available tools.
+## Your process (follow this exact order)
 
-## Mandatory process — follow this order for EVERY item you extract
+### 1. Identify the project
+File: {filename}
+{metadata_section}
 
-1. READ the existing records listed below for the relevant domain.
+Existing projects:
-2. SEARCH for a match by title, topic, or semantic similarity.
+{projects_list}
-3. If a match exists → call the update_* tool with the existing record's id.
-4. If no match exists → call the create_* tool and set isAiSuggested=1.
 
-NEVER call create_* without first checking the existing records.
+Match this file to an existing project using the filename and content clues.
-NEVER duplicate a record that already exists under a different wording.
+If no project matches, {no_match_behavior}.
 
-## Existing records (source of truth)
+### 2. Check existing records
+Once you identify the project, use list_tasks / list_notes / list_timelines
+(filtered by projectId) to see what already exists.
+NEVER create a record that already exists under the same or similar title.
 
-{existing_context}
+### 3. Extract and create / update
+{extraction_rules}
 
-## Context
+### Rules
-
+- Set isAiSuggested=1 on every new record.
-Project: {project_context}
+- Set projectId on every record (use the id from the project list above).
-Domains to extract: {data_types}
+- Update existing records when a match is found by title or topic.
-
+- Do NOT invent data — only extract what is clearly stated in the content.
-{custom_prompt_section}
+- Target entity types: {data_types}.
+{global_rules}
 """
 
 # ── Cloud processing prompt (kept separate for cloud agent) ───────────────
 
-_CLOUD_PROCESSING_PROMPT = """\
+_BATCH_CLOUD_PROCESSING_PROMPT = """\
 You are a data extraction and management assistant for a freelance project
 management tool.
 
@@ -268,9 +225,19 @@ async def _run_agent_with_tools(
     user_message: str,
     tools: list[Any],
     max_steps: int,
+    user_id: str = "",
+    session_id: str = "",
+    langfuse_prompt: Any = None,
+    agent_name: str = "batch-agent",
+    _tool_calls_out: list[str] | None = None,
 ) -> str:
-    """Run an LLM agent with tool-calling, returning the final text response."""
+    """Run an LLM agent with tool-calling, returning the final text response.
-    llm = get_llm()
+
+    If *_tool_calls_out* is provided, the name of every tool called during the
+    run is appended to it (used by the caller to count ``create_*`` calls).
+    """
+    lf = get_langfuse()
+    llm = get_agent_llm(agent_name)
     llm_with_tools = llm.bind_tools(tools)
     messages: list[Any] = [
         SystemMessage(content=system_prompt),
@@ -279,38 +246,83 @@ async def _run_agent_with_tools(
 
     tool_map = {tool_def.name: tool_def for tool_def in tools}
 
-    for _ in range(max_steps):
+    _lf_ctx = langfuse_context(user_id=user_id or None, session_id=session_id or None)
-        response: AIMessage = await llm_with_tools.ainvoke(messages)
+    _lf_ctx.__enter__()
-        messages.append(response)
 
-        if not response.tool_calls:
+    _span_ctx = (
-            return _as_text(response.content)
+        lf.start_as_current_observation(
+            as_type="span",
+            name=agent_name,
+            metadata={"user_id": user_id} if user_id else None,
+            input=user_message,
+        )
+        if lf else None
+    )
+    _span = _span_ctx.__enter__() if _span_ctx else None
 
-        for call in response.tool_calls:
+    try:
-            call_id = str(call.get("id", ""))
+        for _ in range(max_steps):
-            call_name = str(call.get("name", ""))
+            _gen_ctx = (
-            call_args = call.get("args", {})
+                lf.start_as_current_observation(
-            logger.info(
+                    as_type="generation",
-                "agent_runner: tool_call name=%s args=%s",
+                    name=f"{agent_name}-llm",
-                call_name,
+                    model=model_for_agent(agent_name),
-                json.dumps(call_args, ensure_ascii=True)[:800],
+                    prompt=langfuse_prompt,
+                    input=messages,
+                )
+                if lf else None
             )
+            _gen = _gen_ctx.__enter__() if _gen_ctx else None
+            response: AIMessage = await llm_with_tools.ainvoke(messages)
+            if _gen_ctx:
+                _gen.update(output=_as_text(response.content), usage_details=extract_usage(response))
+                _gen_ctx.__exit__(None, None, None)
 
-            tool_fn = tool_map.get(call_name)
+            messages.append(response)
-            if tool_fn is None:
-                tool_output = f"Unknown tool: {call_name}"
-            else:
-                tool_output = await tool_fn.ainvoke(call_args)
 
-            logger.info(
+            if not response.tool_calls:
-                "agent_runner: tool_result name=%s output=%s",
+                final_text = _as_text(response.content)
-                call_name,
+                if _span:
-                str(tool_output)[:200],
+                    _span.update(output=final_text)
-            )
+                return final_text
-            messages.append(ToolMessage(content=str(tool_output), tool_call_id=call["id"]))
 
-    final = await llm.ainvoke(messages)
+            for call in response.tool_calls:
-    return _as_text(final.content)
+                call_id = str(call.get("id", ""))
+                call_name = str(call.get("name", ""))
+                call_args = call.get("args", {})
+                logger.info(
+                    "agent_runner: tool_call name=%s args=%s",
+                    call_name,
+                    json.dumps(call_args, ensure_ascii=True)[:800],
+                )
+
+                if _tool_calls_out is not None:
+                    _tool_calls_out.append(call_name)
+
+                tool_fn = tool_map.get(call_name)
+                if tool_fn is None:
+                    tool_output = f"Unknown tool: {call_name}"
+                else:
+                    tool_output = await tool_fn.ainvoke(call_args)
+
+                logger.info(
+                    "agent_runner: tool_result name=%s output=%s",
+                    call_name,
+                    str(tool_output)[:200],
+                )
+                messages.append(ToolMessage(content=str(tool_output), tool_call_id=call["id"]))
+
+        final = await llm.ainvoke(messages)
+        final_text = _as_text(final.content)
+        if _span:
+            _span.update(output=final_text)
+        return final_text
+    finally:
+        if _span_ctx:
+            _span_ctx.__exit__(None, None, None)
+        _lf_ctx.__exit__(None, None, None)
+        if lf:
+            lf.flush()
 
 
 # ── Tool list builder ─────────────────────────────────────────────────────
@@ -377,7 +389,8 @@ async def _scan_directories(
     for file_path in all_files:
         try:
             meta = await execute_on_client(action="get_file_metadata", data={"path": file_path})
-            modified_at = meta.get("modifiedAt")
+            # FE sends snake_case keys on the wire (toSnakeCase transform)
+            modified_at = meta.get("modified_at") or meta.get("modifiedAt")
             if modified_at is None:
                 filtered.append(file_path)
                 continue
@@ -479,83 +492,66 @@ def _format_entities_for_context(domain: str, rows: list[dict]) -> str:
     return f"Existing {domain}:\n" + "\n".join(lines)
 
 
-# ── Step 1: LLM file classifier ───────────────────────────────────────────
+# ── V2 helper functions ───────────────────────────────────────────────────
 
 
-async def _classify_file(
+def _format_projects(projects: list[dict]) -> str:
-    file_path: str,
+    """Format the project list for the unified system prompt."""
-    file_content: str,
+    if not projects:
-    projects: list[dict],
+        return "  (no projects yet)"
-    config_data_types: list[str],
+    lines: list[str] = []
-) -> tuple[str, list[str], str | None]:
+    for p in projects:
-    """Call the LLM to classify a file by project and relevant domains.
-
-    Returns ``(project_id_or_"new", domains, new_project_name_or_None)``.
-    - ``project_id`` is an existing project UUID, or ``"new"`` when no match found.
-    - ``new_project_name`` is only set when ``project_id == "new"``.
-    Falls back to ``("new", config_data_types, None)`` on any error.
-    """
-    fallback: tuple[str, list[str], str | None] = ("new", list(config_data_types), None)
-
-    if not file_content.strip():
-        return fallback
-
-    valid_project_ids = {p["id"] for p in projects}
-
-    def _fmt_project(p: dict) -> str:
         summary = (p.get("aiSummary") or p.get("ai_summary") or "").strip()
         summary_part = f" — {summary[:100]}" if summary else ""
-        return f"  - id={p['id']} | name={p.get('name', '')} | status={p.get('status', '')}{summary_part}"
+        lines.append(
+            f"  - id={p['id']} | name={p.get('name', '')} | "
+            f"status={p.get('status', '')}{summary_part}"
+        )
+    return "\n".join(lines)
 
-    projects_list = "\n".join(_fmt_project(p) for p in projects) or "  (none yet)"
 
-    domain_definitions = "\n".join(
+def _format_metadata(metadata: dict) -> str:
-        f"  - {d}: {_DOMAIN_DESCRIPTIONS[d]}"
+    """Format preprocessor metadata as a compact context block."""
-        for d in config_data_types
+    if not metadata:
-        if d in _DOMAIN_DESCRIPTIONS
+        return ""
+    parts: list[str] = []
+    for key in ("subject", "from", "to", "date"):
+        if metadata.get(key):
+            parts.append(f"{key.capitalize()}: {metadata[key]}")
+    # any remaining keys
+    for key, val in metadata.items():
+        if key not in ("subject", "from", "to", "date") and val:
+            parts.append(f"{key}: {val}")
+    return "\n".join(parts)
+
+
+def _get_extraction_rules(agent_config: dict, content_type: str) -> str:
+    """Return the extraction_prompt for *content_type* from *agent_config*.
+
+    Falls back to a generic instruction when the type is not configured.
+    """
+    for ct in agent_config.get("content_types", []):
+        if ct.get("id") == content_type:
+            prompt = ct.get("extraction_prompt", "").strip()
+            if prompt:
+                return prompt
+    return (
+        "Extract relevant information as tasks (action items), notes "
+        "(informational content), or timelines (dated events)."
     )
 
-    system = _STEP1_SYSTEM_PROMPT.format(
-        domain_definitions=domain_definitions,
-        projects_list=projects_list,
-    )
 
-    llm = get_llm()
+def _get_no_match_behavior(agent_config: dict) -> str:
-    try:
+    """Derive the 'no project match' instruction from global_rules."""
-        response = await llm.ainvoke([
+    rules = agent_config.get("global_rules", [])
-            SystemMessage(content=system),
+    for rule in rules:
-            HumanMessage(content=f"File: {file_path}\n\nContent:\n{file_content[:4000]}"),
+        lower = rule.lower()
-        ])
+        if "no project" in lower or "no match" in lower or "skip" in lower:
-        raw = _as_text(response.content).strip()
+            return rule
-        # Strip markdown fences if the model wraps the JSON.
+    return "create a new project with a concise name derived from the file content"
-        if raw.startswith("```"):
-            raw = raw.split("```")[1]
-            if raw.startswith("json"):
-                raw = raw[4:]
-        parsed = json.loads(raw.strip())
-        raw_project_id: str = str(parsed.get("project_id") or "new")
-        # Reject hallucinated UUIDs — only accept ids that exist in the fetched list.
-        project_id = raw_project_id if raw_project_id in valid_project_ids else "new"
-        new_project_name: str | None = (
-            str(parsed["new_project_name"]).strip() or None
-            if project_id == "new" and parsed.get("new_project_name")
-            else None
-        )
-        domains: list[str] = [
-            d for d in parsed.get("domains", [])
-            if d in config_data_types
-        ]
-        if not domains:
-            domains = list(config_data_types)
-        return project_id, domains, new_project_name
-    except Exception as exc:
-        logger.warning(
-            "agent_runner: step1 classification failed for %r: %s", file_path, exc
-        )
-        return fallback
 
 
-# ── Local agent runner (two-step per file) ────────────────────────────────
+# ── Local agent runner (V2 — unified per-file flow) ───────────────────────
 
 
 async def run_local_agent(
@@ -565,16 +561,17 @@ async def run_local_agent(
     device_mgr: DeviceConnectionManager,
     run_context: dict | None = None,
 ) -> None:
-    """Execute a local directory agent run using a two-step approach per file.
+    """Execute a local directory agent run — V2 unified flow.
 
-    Step 1 — Classification (code + 1 LLM call per file, no tools):
+    Phase A — Detect + Preprocess (zero LLM, per file):
-        Code scans directories and fetches all projects via WS.
+        Python detects the content type from filename + content patterns and
-        For each file, LLM identifies the project and relevant domains.
+        runs the appropriate handler (e.g. email_html) to produce clean text
+        and structured metadata.
 
-    Step 2 — Processing (code + 1 LLM call per file, with tools):
+    Phase B — Single LLM call with tools (per file):
-        Code fetches existing entities for the identified project/domains.
+        One LLM call handles project identification, duplicate checking, and
-        LLM receives file content + existing entities in context and uses
+        record creation/update.  ``create_*`` tool calls are counted to
-        tools to update existing records or create new ones.
+        produce the accurate ``items_created`` metric.
     """
     run_id = run_log.id
     agent_id = (run_context or {}).get("agent_id") or config.id
@@ -609,16 +606,11 @@ async def run_local_agent(
     errors: list[str] = []
     items_processed = 0
     items_created = 0
-
+    agent_config: dict = config.agent_config or {}
-    custom_section = (
+    processing_tools = _build_processing_tools(config.data_types)
-        f"User instructions:\n{config.prompt_template}"
-        if config.prompt_template
-        else ""
-    )
 
     try:
         # ── Code: scan directories ───────────────────────────────────
-        logger.info("agent_runner: run=%s scanning directories user=%s", run_id, user_id)
         file_paths = await _scan_directories(
             paths=config.directory_paths,
             extensions=config.file_extensions or [],
@@ -634,108 +626,89 @@ async def run_local_agent(
 
         # ── Code: fetch all projects once ────────────────────────────
         projects = await _fetch_projects()
+        projects_block = _format_projects(projects)
+
+        # Prompt template + Langfuse version linking (hot-swappable from UI).
+        unified_template, prompt_obj = get_prompt_or_fallback(
+            "unified_processing", _UNIFIED_PROCESSING_PROMPT
+        )
 
         for file_path in file_paths:
             try:
-                # Read file content via code.
+                # ── Phase A: read + detect + preprocess ─────────────
                 file_result = await execute_on_client(
                     action="read_file_content", data={"path": file_path}
                 )
-                file_content: str = file_result.get("content", "")
+                raw_content: str = file_result.get("content", "")
-                if not file_content:
+                if not raw_content.strip():
-                    logger.debug("agent_runner: run=%s skipping empty file %r", run_id, file_path)
+                    logger.debug(
+                        "agent_runner: run=%s skipping empty file %r", run_id, file_path
+                    )
                     continue
 
                 items_processed += 1
+                filename = os.path.basename(file_path)
+                content_type = detect_content_type(filename, raw_content)
+                preprocessed = preprocess(content_type, raw_content)
 
-                # Step 1 — classify file.
-                project_id, domains, new_project_name = await _classify_file(
-                    file_path=file_path,
-                    file_content=file_content,
-                    projects=projects,
-                    config_data_types=config.data_types,
-                )
                 logger.info(
-                    "agent_runner: run=%s file=%r → project=%s new_name=%r domains=%s",
+                    "agent_runner: run=%s file=%r content_type=%s clean_len=%d",
-                    run_id,
+                    run_id, file_path, content_type, len(preprocessed.clean_text),
-                    file_path,
-                    project_id,
-                    new_project_name,
-                    domains,
                 )
 
-                # Step 2 — resolve project_id via CODE, then fetch entities.
+                # ── Phase B: single LLM call ─────────────────────────
-                # Project creation is NEVER delegated to the Step 2 LLM.
+                extraction_rules = _get_extraction_rules(agent_config, content_type)
-                if project_id == "new":
+                no_match_behavior = _get_no_match_behavior(agent_config)
-                    proj_name = new_project_name or "Untitled Project"
+                global_rules_lines = "\n".join(
-                    try:
+                    f"- {r}" for r in agent_config.get("global_rules", [])
-                        proj_result = await execute_on_client(
+                )
-                            action="insert",
+                metadata_section = _format_metadata(preprocessed.metadata)
-                            table="projects",
-                            data={"name": proj_name, "clientId": None},
-                        )
-                        created = proj_result.get("row", {})
-                        effective_project_id = created.get("id", "standalone")
-                        # Add to local list so subsequent files can match it.
-                        if "id" in created:
-                            projects.append(created)
-                        logger.info(
-                            "agent_runner: run=%s created project %r id=%s",
-                            run_id, proj_name, effective_project_id,
-                        )
-                    except Exception as exc:
-                        logger.warning(
-                            "agent_runner: run=%s failed to create project %r: %s",
-                            run_id, proj_name, exc,
-                        )
-                        effective_project_id = "standalone"
-                        proj_name = "unknown"
-                    project_context = (
-                        f"Project: {proj_name} (id: {effective_project_id}). "
-                        "Always set projectId to this id on every record you create."
-                    )
-                else:
-                    effective_project_id = project_id
-                    proj = next((p for p in projects if p["id"] == project_id), None)
-                    proj_name = proj.get("name", project_id) if proj else project_id
-                    project_context = (
-                        f"Project: {proj_name} (id: {project_id}). "
-                        "Always set projectId to this id on every record you create."
-                    )
 
-                # "projects" domain is never passed to Step 2 — handled above in code.
+                system_prompt = compile_prompt(
-                domains = [d for d in domains if d != "projects"]
+                    unified_template,
-
+                    prompt_obj,
-                existing_blocks: list[str] = []
+                    filename=filename,
-                for domain in domains:
+                    metadata_section=metadata_section,
-                    rows = await _fetch_domain_entities(domain, effective_project_id)
+                    projects_list=projects_block,
-                    existing_blocks.append(_format_entities_for_context(domain, rows))
+                    no_match_behavior=no_match_behavior,
-
+                    extraction_rules=extraction_rules,
-                existing_context = "\n\n".join(existing_blocks)
+                    global_rules=global_rules_lines,
-
+                    data_types=", ".join(config.data_types),
-                system_prompt = _PROCESSING_SYSTEM_PROMPT.format(
-                    existing_context=existing_context,
-                    project_context=project_context,
-                    data_types=", ".join(domains),
-                    custom_prompt_section=custom_section,
                 )
 
-                processing_tools = _build_processing_tools(domains)
+                user_message = (
+                    f"Process this file and extract relevant information.\n\n"
+                    f"File: {file_path}\n\n"
+                    f"Content:\n{preprocessed.clean_text}"
+                )
 
+                file_tool_calls: list[str] = []
                 result_text = await _run_agent_with_tools(
                     system_prompt=system_prompt,
-                    user_message=(
+                    user_message=user_message,
-                        f"Process this file and extract relevant information.\n\n"
-                        f"File: {file_path}\n\nContent:\n{file_content}"
-                    ),
                     tools=processing_tools,
                     max_steps=_MAX_PROCESSING_STEPS,
+                    user_id=user_id,
+                    session_id=run_id,
+                    langfuse_prompt=prompt_obj,
+                    agent_name="unified-processor",
+                    _tool_calls_out=file_tool_calls,
                 )
+
+                file_created = sum(
+                    1 for name in file_tool_calls if name.startswith("create_")
+                )
+                items_created += file_created
+
+                # Refresh project list when a project was created so
+                # subsequent files see it in the prompt context.
+                if "create_project" in file_tool_calls:
+                    projects = await _fetch_projects()
+                    projects_block = _format_projects(projects)
+
                 logger.info(
-                    "agent_runner: run=%s file=%r result=%s",
+                    "agent_runner: run=%s file=%r created=%d result=%s",
-                    run_id,
+                    run_id, file_path, file_created, result_text[:200],
-                    file_path,
-                    result_text[:200],
                 )
 
             except Exception as exc:
@@ -767,10 +740,11 @@ async def run_local_agent(
         errors=errors,
     )
     logger.info(
-        "agent_runner: run=%s done status=%s processed=%d errors=%d",
+        "agent_runner: run=%s done status=%s processed=%d created=%d errors=%d",
         run_id,
         final_status,
         items_processed,
+        items_created,
         len(errors),
     )
 
@@ -928,7 +902,12 @@ async def run_cloud_agent(
                 continue
             items_processed += 1
 
-            processing_prompt = _CLOUD_PROCESSING_PROMPT.format(
+            cloud_template, cloud_prompt_obj = get_prompt_or_fallback(
+                "batch_cloud_processing", _BATCH_CLOUD_PROCESSING_PROMPT
+            )
+            processing_prompt = compile_prompt(
+                cloud_template,
+                cloud_prompt_obj,
                 data_types=", ".join(config.data_types),
                 project_context="Determine the appropriate project from the message context.",
                 file_list=f"Message from {config.provider} (id: {msg.id})",
@@ -941,6 +920,10 @@ async def run_cloud_agent(
                     user_message=f"Process this message content:\n\n{content_text[:8000]}",
                     tools=processing_tools,
                     max_steps=_MAX_PROCESSING_STEPS,
+                    user_id=user_id,
+                    session_id=run_id,
+                    langfuse_prompt=cloud_prompt_obj,
+                    agent_name="cloud-processor",
                 )
             except Exception as exc:
                 errors.append(f"LLM processing error for message {msg.id!r}: {exc}")

app/core/deep_agent.py

@@ -16,7 +16,8 @@ from app.agents.note_agent import NOTE_TOOLS
 from app.agents.project_agent import PROJECT_TOOLS
 from app.agents.task_agent import TASK_TOOLS
 from app.agents.timeline_agent import TIMELINE_TOOLS
-from app.core.llm import get_llm
+from app.core.langfuse_client import extract_usage, get_langfuse, get_prompt_or_fallback, langfuse_context
+from app.core.llm import get_agent_llm, model_for_agent
 from app.core.memory_middleware import MemoryMiddleware
 from app.core.ws_context import clear_tool_result_collector, execute_on_client, set_tool_result_collector
 from app.db import async_session
@@ -26,7 +27,67 @@ logger = logging.getLogger(__name__)
 FloatingDomainType = Literal["task", "timeline", "project", "node"]
 FloatingDomainSection = Literal["task", "timeline", "note"]
 
-_HOME_SINGLE_AGENT_SYSTEM = (
+# Mapping of core-memory language values to natural-language names for prompts.
+_LANGUAGE_NAMES: dict[str, str] = {
+    "en": "English", "it": "Italian", "es": "Spanish",
+    "fr": "French", "de": "German",
+    "english": "English", "italian": "Italian", "italiano": "Italian",
+    "spanish": "Spanish", "español": "Spanish",
+    "french": "French", "français": "French",
+    "german": "German", "deutsch": "German",
+}
+
+
+def _language_instruction(context: dict[str, Any]) -> str:
+    """Return a system-prompt suffix that tells the LLM to respond in the user's language.
+
+    Returns an empty string when the language is English or unknown — saves tokens.
+    """
+    core = context.get("core_memory") or {}
+    raw = (core.get("language") or "").strip().lower()
+    if not raw:
+        return ""
+    lang = _LANGUAGE_NAMES.get(raw, raw.title())  # best-effort capitalisation
+    if lang.lower() == "english":
+        return ""
+    return (
+        f"\n\nIMPORTANT: Always respond in {lang}. "
+        f"All your output text must be written in {lang}."
+    )
+
+def _proactive_hints_injection(context: dict[str, Any]) -> str:
+    """Return a system-prompt paragraph listing proactive behavioral hints.
+
+    Returns empty string when no hints or confidence below threshold.
+    Capped at 600 chars.
+    """
+    hints: list[str] = context.get("proactive_hints") or []
+    if not hints:
+        return ""
+    body = "\n".join(f"- {h}" for h in hints)
+    section = f"\n\nI noticed (behavioral patterns):\n{body}"
+    if len(section) > 600:
+        section = section[:597] + "..."
+    return section
+
+
+def _relational_memory_injection(context: dict[str, Any]) -> str:
+    """Return a system-prompt paragraph listing known people/projects from relational memory.
+
+    Returns empty string when no relational rows or tier is Free.
+    Capped at 800 chars to control token spend.
+    """
+    relations: list[str] = context.get("relational_memory") or []
+    if not relations:
+        return ""
+    body = "\n".join(f"- {r}" for r in relations)
+    section = f"\n\nKnown people & projects:\n{body}"
+    if len(section) > 800:
+        section = section[:797] + "..."
+    return section
+
+
+_HOME_SYSTEM_PROMPT = (
     "You are the home assistant with direct access to all tools: tasks, projects, notes, timelines, and memory tools. "
     "Always use tools for factual data retrieval before answering. "
     "When the user asks to remember, forget, or update what you know about them, use memory tools. "
@@ -39,7 +100,7 @@ _HOME_SINGLE_AGENT_SYSTEM = (
     "For upcoming tasks, after tag lines add a short recommendation based on due date and priority."
 )
 
-_FLOATING_SINGLE_AGENT_SYSTEM = (
+_FLOATING_SYSTEM_PROMPT = (
     "You are the floating assistant with direct access to all tools: tasks, projects, notes, timelines, and memory tools. "
     "Stay focused on the floating scope in context.scope and answer concisely. "
     "Return plain text only. Do not output XML/HTML-like tags such as <task>, <project>, <note>, <timeline>, or any bracketed id tag wrappers. "
@@ -48,7 +109,7 @@ _FLOATING_SINGLE_AGENT_SYSTEM = (
     "If context.context.resolved_project_id exists, use it as project_id for scoped list calls. "
 )
 
-_FLOATING_DOMAIN_CLASSIFIER_SYSTEM = (
+_FLOATING_DOMAIN_CLASSIFIER_PROMPT = (
     "You are a strict domain classifier for websocket floating requests. "
     "Return ONLY a JSON object with keys: type, id, section. "
     "Allowed type values: task, timeline, project, node. "
@@ -147,6 +208,15 @@ def _trace_id_from_context(context: dict[str, Any]) -> str | None:
     return None
 
 
+def _session_id_from_context(context: dict[str, Any]) -> str | None:
+    debug = context.get("_debug")
+    if isinstance(debug, dict):
+        session_id = debug.get("session_id")
+        if isinstance(session_id, str) and session_id:
+            return session_id
+    return None
+
+
 def _context_for_model(context: dict[str, Any]) -> dict[str, Any]:
     sanitized = dict(context)
     sanitized.pop("_debug", None)
@@ -535,18 +605,39 @@ async def _infer_floating_domain(message: str, context: dict[str, Any]) -> dict[
     }
 
     try:
-        llm = get_llm()
+        llm = get_agent_llm("classifier")
-        response = await llm.ainvoke(
+        classifier_messages = [
-            [
+            SystemMessage(content=_FLOATING_DOMAIN_CLASSIFIER_PROMPT),
-                SystemMessage(content=_FLOATING_DOMAIN_CLASSIFIER_SYSTEM),
+            HumanMessage(
-                HumanMessage(
+                content=(
-                    content=(
+                    f"Message:\n{message}\n\n"
-                        f"Message:\n{message}\n\n"
+                    f"Context:\n{json.dumps(classifier_context, ensure_ascii=True)}"
-                        f"Context:\n{json.dumps(classifier_context, ensure_ascii=True)}"
+                )
-                    )
+            ),
-                ),
+        ]
-            ]
+        lf = get_langfuse()
+        _, classifier_prompt_obj = get_prompt_or_fallback(
+            "floating_domain_classifier", _FLOATING_DOMAIN_CLASSIFIER_PROMPT
         )
+
+        # Extract user/session from context for Langfuse attribution
+        _debug = context.get("_debug") if isinstance(context, dict) else None
+        _lf_user = (_debug or {}).get("user_id") if isinstance(_debug, dict) else None
+        _lf_session = (_debug or {}).get("session_id") if isinstance(_debug, dict) else None
+
+        with langfuse_context(user_id=_lf_user, session_id=_lf_session):
+            if lf:
+                with lf.start_as_current_observation(
+                    as_type="generation",
+                    name="floating-classifier",
+                    model=model_for_agent("classifier"),
+                    prompt=classifier_prompt_obj,
+                    input=classifier_messages,
+                ) as gen:
+                    response = await llm.ainvoke(classifier_messages)
+                    gen.update(output=_as_text(response.content), usage_details=extract_usage(response))
+            else:
+                response = await llm.ainvoke(classifier_messages)
         parsed = _parse_json_object(_as_text(response.content))
         if parsed is not None:
             domain = _normalize_domain_payload(parsed, project_id)
@@ -571,9 +662,13 @@ async def _run_single_agent(
     message: str,
     context: dict[str, Any],
     max_steps: int = 6,
+    langfuse_prompt: Any = None,
+    agent_name: str = "agent",
 ) -> str:
     trace_id = _trace_id_from_context(context)
-    llm = get_llm()
+    session_id = _session_id_from_context(context)
+    lf = get_langfuse()
+    llm = get_agent_llm(agent_name)
     tools = _all_tools_for_user(user_id, trace_id)
     model_context = _context_for_model(context)
     logger.info("deep_agent: run_single_agent_start trace=%s user=%s", trace_id or "-", user_id)
@@ -591,9 +686,39 @@ async def _run_single_agent(
     tool_calls_count = 0
     collected: list[dict[str, Any]] = []
     set_tool_result_collector(collected)
+
+    _lf_ctx = langfuse_context(user_id=user_id, session_id=session_id)
+    _lf_ctx.__enter__()
+
+    _span_ctx = (
+        lf.start_as_current_observation(
+            as_type="span",
+            name=agent_name,
+            metadata={"user_id": user_id, "session_id": trace_id},
+            input=message,
+        )
+        if lf else None
+    )
+    _span = _span_ctx.__enter__() if _span_ctx else None
+
     try:
         for _ in range(max_steps):
+            _gen_ctx = (
+                lf.start_as_current_observation(
+                    as_type="generation",
+                    name=f"{agent_name}-llm",
+                    model=model_for_agent(agent_name),
+                    prompt=langfuse_prompt,
+                    input=messages,
+                )
+                if lf else None
+            )
+            _gen = _gen_ctx.__enter__() if _gen_ctx else None
             response: AIMessage = await llm_with_tools.ainvoke(messages)
+            if _gen_ctx:
+                _gen.update(output=_as_text(response.content), usage_details=extract_usage(response))
+                _gen_ctx.__exit__(None, None, None)
+
             messages.append(response)
 
             if not response.tool_calls:
@@ -605,6 +730,8 @@ async def _run_single_agent(
                     tool_calls_count,
                     len(final_text),
                 )
+                if _span:
+                    _span.update(output=final_text)
                 return final_text
 
             tool_map = {tool_def.name: tool_def for tool_def in tools}
@@ -644,9 +771,16 @@ async def _run_single_agent(
             tool_calls_count,
             len(final_text),
         )
+        if _span:
+            _span.update(output=final_text)
         return final_text
     finally:
         clear_tool_result_collector()
+        if _span_ctx:
+            _span_ctx.__exit__(None, None, None)
+        _lf_ctx.__exit__(None, None, None)
+        if lf:
+            lf.flush()
 
 
 async def _run_single_agent_stream(
@@ -656,9 +790,13 @@ async def _run_single_agent_stream(
     message: str,
     context: dict[str, Any],
     max_steps: int = 6,
+    langfuse_prompt: Any = None,
+    agent_name: str = "agent",
 ) -> AsyncGenerator[tuple[str, Any], None]:
     trace_id = _trace_id_from_context(context)
-    llm = get_llm()
+    session_id = _session_id_from_context(context)
+    lf = get_langfuse()
+    llm = get_agent_llm(agent_name)
     tools = _all_tools_for_user(user_id, trace_id)
     model_context = _context_for_model(context)
     logger.info("deep_agent: run_single_agent_stream_start trace=%s user=%s", trace_id or "-", user_id)
@@ -677,9 +815,40 @@ async def _run_single_agent_stream(
     streamed_chars = 0
     collected: list[dict[str, Any]] = []
     set_tool_result_collector(collected)
+
+    _lf_ctx = langfuse_context(user_id=user_id, session_id=session_id)
+    _lf_ctx.__enter__()
+
+    _span_ctx = (
+        lf.start_as_current_observation(
+            as_type="span",
+            name=f"{agent_name}-stream",
+            metadata={"user_id": user_id, "session_id": trace_id},
+            input=message,
+        )
+        if lf else None
+    )
+    _span = _span_ctx.__enter__() if _span_ctx else None
+    streamed_text: list[str] = []
+
     try:
         for _ in range(max_steps):
+            _gen_ctx = (
+                lf.start_as_current_observation(
+                    as_type="generation",
+                    name=f"{agent_name}-llm",
+                    model=model_for_agent(agent_name),
+                    prompt=langfuse_prompt,
+                    input=messages,
+                )
+                if lf else None
+            )
+            _gen = _gen_ctx.__enter__() if _gen_ctx else None
             response: AIMessage = await llm_with_tools.ainvoke(messages)
+            if _gen_ctx:
+                _gen.update(output=_as_text(response.content), usage_details=extract_usage(response))
+                _gen_ctx.__exit__(None, None, None)
+
             messages.append(response)
 
             if not response.tool_calls:
@@ -688,6 +857,7 @@ async def _run_single_agent_stream(
                     token = _as_text(getattr(chunk, "content", ""))
                     if token:
                         streamed_chars += len(token)
+                        streamed_text.append(token)
                         emitted_any = True
                         yield "token", token
 
@@ -696,6 +866,7 @@ async def _run_single_agent_stream(
                     fallback_text = _as_text(response.content)
                     if fallback_text:
                         streamed_chars += len(fallback_text)
+                        streamed_text.append(fallback_text)
                         yield "token", fallback_text
                 logger.info(
                     "deep_agent: run_single_agent_stream_end trace=%s user=%s tool_calls=%d response_chars=%d",
@@ -704,6 +875,8 @@ async def _run_single_agent_stream(
                     tool_calls_count,
                     streamed_chars,
                 )
+                if _span:
+                    _span.update(output="".join(streamed_text))
                 return
 
             tool_map = {tool_def.name: tool_def for tool_def in tools}
@@ -738,6 +911,7 @@ async def _run_single_agent_stream(
             token = _as_text(getattr(chunk, "content", ""))
             if token:
                 streamed_chars += len(token)
+                streamed_text.append(token)
                 yield "token", token
         logger.info(
             "deep_agent: run_single_agent_stream_end trace=%s user=%s tool_calls=%d response_chars=%d fallback=1",
@@ -746,17 +920,32 @@ async def _run_single_agent_stream(
             tool_calls_count,
             streamed_chars,
         )
+        if _span:
+            _span.update(output="".join(streamed_text))
     finally:
         clear_tool_result_collector()
+        if _span_ctx:
+            _span_ctx.__exit__(None, None, None)
+        _lf_ctx.__exit__(None, None, None)
+        if lf:
+            lf.flush()
 
 
 async def run_home(user_id: str, message: str, context: dict[str, Any]) -> str:
     prepared_context = await _prepare_context(message, context)
+    system_prompt, langfuse_prompt = get_prompt_or_fallback(
+        "home_system", _HOME_SYSTEM_PROMPT
+    )
+    system_prompt += _relational_memory_injection(context)
+    system_prompt += _proactive_hints_injection(context)
+    system_prompt += _language_instruction(context)
     response = await _run_single_agent(
         user_id=user_id,
-        system_prompt=_HOME_SINGLE_AGENT_SYSTEM,
+        system_prompt=system_prompt,
         message=message,
         context=prepared_context,
+        langfuse_prompt=langfuse_prompt,
+        agent_name="home-agent",
     )
     return _normalize_tagged_list_lines(response, message)
 
@@ -764,11 +953,19 @@ async def run_home(user_id: str, message: str, context: dict[str, Any]) -> str:
 async def run_floating(user_id: str, message: str, context: dict[str, Any]) -> tuple[str, dict[str, str | None]]:
     prepared_context = await _prepare_context(message, context)
     domain = await _infer_floating_domain(message, prepared_context)
+    system_prompt, langfuse_prompt = get_prompt_or_fallback(
+        "floating_system", _FLOATING_SYSTEM_PROMPT
+    )
+    system_prompt += _relational_memory_injection(context)
+    system_prompt += _proactive_hints_injection(context)
+    system_prompt += _language_instruction(context)
     response = await _run_single_agent(
         user_id=user_id,
-        system_prompt=_FLOATING_SINGLE_AGENT_SYSTEM,
+        system_prompt=system_prompt,
         message=message,
         context=prepared_context,
+        langfuse_prompt=langfuse_prompt,
+        agent_name="floating-agent",
     )
     sanitized = _strip_floating_markup(response)
     if not sanitized and response:
@@ -782,12 +979,20 @@ async def run_home_stream(
     context: dict[str, Any],
 ) -> AsyncGenerator[tuple[str, Any], None]:
     prepared_context = await _prepare_context(message, context)
+    system_prompt, langfuse_prompt = get_prompt_or_fallback(
+        "home_system", _HOME_SYSTEM_PROMPT
+    )
+    system_prompt += _relational_memory_injection(context)
+    system_prompt += _proactive_hints_injection(context)
+    system_prompt += _language_instruction(context)
     text_chunks: list[str] = []
     async for event in _run_single_agent_stream(
         user_id=user_id,
-        system_prompt=_HOME_SINGLE_AGENT_SYSTEM,
+        system_prompt=system_prompt,
         message=message,
         context=prepared_context,
+        langfuse_prompt=langfuse_prompt,
+        agent_name="home-agent",
     ):
         event_type, data = event
         if event_type != "token":
@@ -809,14 +1014,22 @@ async def run_floating_stream(
     domain = await _infer_floating_domain(message, prepared_context)
     yield "floating_domain", domain
 
+    system_prompt, langfuse_prompt = get_prompt_or_fallback(
+        "floating_system", _FLOATING_SYSTEM_PROMPT
+    )
+    system_prompt += _relational_memory_injection(context)
+    system_prompt += _proactive_hints_injection(context)
+    system_prompt += _language_instruction(context)
     sanitizer = _FloatingStreamSanitizer()
     emitted_sanitized = False
     raw_chunks: list[str] = []
     async for event in _run_single_agent_stream(
         user_id=user_id,
-        system_prompt=_FLOATING_SINGLE_AGENT_SYSTEM,
+        system_prompt=system_prompt,
         message=message,
         context=prepared_context,
+        langfuse_prompt=langfuse_prompt,
+        agent_name="floating-agent",
     ):
         event_type, data = event
         if event_type != "token":

app/core/embeddings.py

@@ -0,0 +1,34 @@
+"""OpenAI embedding helper for associative memory tier.
+
+Single public function: ``embed_text(text) -> list[float] | None``.
+Returns None on any failure — callers must implement a keyword fallback.
+Never raises; all exceptions are logged as warnings.
+"""
+
+from __future__ import annotations
+
+import logging
+
+from openai import AsyncOpenAI
+
+logger = logging.getLogger(__name__)
+
+_MAX_INPUT_CHARS = 8000
+_EMBEDDING_MODEL = "text-embedding-3-small"
+
+
+async def embed_text(text: str) -> list[float] | None:
+    """Call OpenAI text-embedding-3-small. Return None on failure (caller falls back to keyword)."""
+    try:
+        client = AsyncOpenAI()
+        truncated = text[:_MAX_INPUT_CHARS]
+        response = await client.embeddings.create(
+            input=truncated,
+            model=_EMBEDDING_MODEL,
+        )
+        result: list[float] = response.data[0].embedding
+        logger.debug("embeddings: embed_text dims=%d", len(result))
+        return result
+    except Exception as exc:
+        logger.warning("embeddings: embed_text failed: %s", exc)
+        return None

app/core/langfuse_client.py

@@ -0,0 +1,190 @@
+"""Langfuse observability — singleton client and prompt helpers.
+
+If LANGFUSE_SECRET_KEY / LANGFUSE_PUBLIC_KEY are not set,
+all helpers are no-ops so the app works without Langfuse configured.
+
+Usage
+-----
+Tracing::
+
+    from app.core.langfuse_client import get_langfuse
+
+    lf = get_langfuse()
+    if lf:
+        with lf.start_as_current_observation(as_type="span", name="my-agent") as span:
+            span.update(input=user_message)
+            # ... do work ...
+            span.update(output=result)
+        lf.flush()
+
+Prompt management::
+
+    from app.core.langfuse_client import get_prompt_or_fallback
+
+    text, prompt_obj = get_prompt_or_fallback("home_system", FALLBACK_PROMPT)
+    # Use text as the system prompt; pass prompt_obj to generations for linking.
+
+Linking a prompt to a generation::
+
+    with lf.start_as_current_observation(
+        as_type="generation",
+        name="llm-call",
+        model="gpt-4o",
+        prompt=prompt_obj,   # links generation → prompt version in the UI
+        input=messages,
+    ) as gen:
+        response = await llm.ainvoke(messages)
+        gen.update(output=response.content, usage=_usage(response))
+"""
+
+from __future__ import annotations
+
+import hashlib
+import logging
+from contextlib import contextmanager
+from typing import Any, Generator
+
+logger = logging.getLogger(__name__)
+
+_client: Any = None
+_initialized: bool = False
+
+
+def get_langfuse() -> Any | None:
+    """Return the Langfuse singleton, or ``None`` when not configured."""
+    global _client, _initialized
+    if _initialized:
+        return _client
+    _initialized = True
+
+    from app.config.settings import settings  # local import to avoid circular deps
+
+    if not settings.LANGFUSE_SECRET_KEY or not settings.LANGFUSE_PUBLIC_KEY:
+        logger.debug("langfuse: not configured — observability disabled")
+        return None
+
+    try:
+        from langfuse import Langfuse
+
+        _client = Langfuse(
+            secret_key=settings.LANGFUSE_SECRET_KEY,
+            public_key=settings.LANGFUSE_PUBLIC_KEY,
+            host=settings.LANGFUSE_BASE_URL,
+        )
+        logger.info("langfuse: client initialized host=%s", settings.LANGFUSE_BASE_URL)
+    except Exception as exc:
+        logger.warning("langfuse: failed to initialize: %s", exc)
+        _client = None
+
+    return _client
+
+
+def get_prompt_or_fallback(name: str, fallback: str) -> tuple[str, Any]:
+    """Fetch a text prompt from Langfuse; fall back to ``fallback`` on any error.
+
+    Returns ``(raw_template, prompt_obj_or_None)``.
+
+    * ``raw_template`` — the uncompiled template string.  Do NOT call ``.format()``
+      on it directly; use :func:`compile_prompt` instead so the correct variable
+      syntax is applied (``{{var}}`` for Langfuse, ``{var}`` for the fallback).
+    * ``prompt_obj`` — the Langfuse prompt object, or ``None`` when Langfuse is
+      unavailable / the fetch failed.  Pass this to generation observations so
+      Langfuse links the generation to the exact prompt version in the UI.
+    """
+    lf = get_langfuse()
+    if lf is None:
+        return fallback, None
+
+    try:
+        prompt = lf.get_prompt(name, label="production", fallback=fallback)
+        # For text-type prompts .prompt holds the raw template string.
+        raw = prompt.prompt if hasattr(prompt, "prompt") and isinstance(prompt.prompt, str) else fallback
+        return raw, prompt
+    except Exception as exc:
+        logger.warning("langfuse: get_prompt %r failed: %s — using fallback", name, exc)
+        return fallback, None
+
+
+def compile_prompt(template: str, prompt_obj: Any, **variables: Any) -> str:
+    """Compile *template* with *variables*, choosing the right syntax.
+
+    * When *prompt_obj* is a real Langfuse prompt object, calls
+      ``prompt_obj.compile(**variables)`` which handles ``{{variable}}``
+      substitution as defined in the Langfuse UI.
+    * When *prompt_obj* is ``None`` (Langfuse unavailable or fetch failed),
+      falls back to ``template.format(**variables)`` which handles the
+      ``{variable}`` syntax used in the hardcoded fallback strings.
+
+    This keeps callers oblivious to which syntax is in use.
+    """
+    if prompt_obj is not None:
+        try:
+            compiled = prompt_obj.compile(**variables)
+            # compile() returns a string for text prompts.
+            if isinstance(compiled, str):
+                return compiled
+            # Chat prompts return a list of dicts — join text parts.
+            if isinstance(compiled, list):
+                return "\n".join(
+                    m.get("content", "") for m in compiled if isinstance(m, dict)
+                )
+        except Exception as exc:
+            logger.warning(
+                "langfuse: compile failed for prompt %r: %s — falling back to .format()",
+                getattr(prompt_obj, "name", "?"),
+                exc,
+            )
+    return template.format(**variables)
+
+
+def extract_usage(response: Any) -> dict[str, int]:
+    """Extract token usage from a LangChain AI message into Langfuse format."""
+    meta = getattr(response, "usage_metadata", None)
+    if not meta:
+        return {}
+    return {
+        "input": int(meta.get("input_tokens", 0)),
+        "output": int(meta.get("output_tokens", 0)),
+        "total": int(meta.get("total_tokens", 0)),
+    }
+
+
+def hash_user_id(user_id: str) -> str:
+    """Return a SHA-256 hash of *user_id* for use as Langfuse ``user_id``.
+
+    This avoids sending raw database UUIDs to external observability services
+    while still providing a stable, deterministic identifier for per-user
+    metrics in the Langfuse dashboard.
+    """
+    return hashlib.sha256(user_id.encode()).hexdigest()
+
+
+@contextmanager
+def langfuse_context(
+    user_id: str | None = None,
+    session_id: str | None = None,
+) -> Generator[None, None, None]:
+    """Propagate ``user_id`` (hashed) and ``session_id`` to all Langfuse observations.
+
+    No-op when Langfuse is not configured or parameters are empty.
+    """
+    lf = get_langfuse()
+    if lf is None or (not user_id and not session_id):
+        yield
+        return
+
+    try:
+        from langfuse import propagate_attributes
+    except ImportError:
+        logger.debug("langfuse: propagate_attributes not available — skipping context")
+        yield
+        return
+
+    attrs: dict[str, str] = {}
+    if user_id:
+        attrs["user_id"] = hash_user_id(user_id)
+    if session_id:
+        attrs["session_id"] = session_id
+
+    with propagate_attributes(**attrs):
+        yield

app/core/llm.py

@@ -1,6 +1,6 @@
 """LLM factory — centralised model instantiation via LiteLLM.
 
-Every agent and the orchestrator call ``get_llm()`` or ``get_router_llm()``
+Every agent and the orchestrator call ``get_llm()``
 instead of directly constructing a provider-specific class.  The model string
 follows the `LiteLLM model naming convention
 <https://docs.litellm.ai/docs/providers>`_:
@@ -11,7 +11,7 @@ follows the `LiteLLM model naming convention
 * Ollama:     ``ollama/llama3``
 * Bedrock:    ``bedrock/anthropic.claude-v2``
 
-Switch providers by changing **LLM_MODEL** / **LLM_ROUTER_MODEL** in ``.env``
+Switch providers by changing **LLM_MODEL** in ``.env``
 — no code changes required.
 """
 
@@ -19,6 +19,7 @@ from __future__ import annotations
 
 import os
 import warnings
+from collections.abc import Callable
 
 from openai import AsyncOpenAI
 import litellm
@@ -95,12 +96,36 @@ def get_llm(
     )
 
 
-def get_router_llm(
+_AGENT_MODEL_SETTINGS: dict[str, Callable[[], str]] = {
+    "classifier":          lambda: settings.LLM_MODEL_CLASSIFIER or settings.LLM_MODEL,
+    "home-agent":          lambda: settings.LLM_MODEL_HOME_AGENT or settings.LLM_MODEL,
+    "floating-agent":      lambda: settings.LLM_MODEL_FLOATING_AGENT or settings.LLM_MODEL,
+    "unified-processor":   lambda: settings.LLM_MODEL_UNIFIED_PROCESSOR or settings.LLM_MODEL,
+    "cloud-processor":     lambda: settings.LLM_MODEL_CLOUD_PROCESSOR or settings.LLM_MODEL,
+    "setup":               lambda: settings.LLM_MODEL_SETUP_AGENT or settings.LLM_MODEL,
+    "memory-extractor":    lambda: settings.LLM_MODEL_MEMORY_EXTRACTOR or "gpt-4o-mini",
+    "memory-miner":        lambda: settings.LLM_MODEL_MEMORY_MINER or "gpt-4o-mini",
+    "memory-auditor":      lambda: settings.LLM_MODEL_MEMORY_AUDITOR or settings.LLM_MODEL,
+}
+
+
+def model_for_agent(agent_name: str) -> str:
+    """Return the resolved model string for *agent_name* (for Langfuse tracking)."""
+    return _AGENT_MODEL_SETTINGS.get(agent_name, lambda: settings.LLM_MODEL)()
+
+
+def get_agent_llm(
+    agent_name: str,
     *,
     temperature: float = 0,
 ) -> ChatOpenAI | ChatLiteLLM:
-    """Return the lighter model used for intent classification / routing."""
+    """Return an LLM configured for *agent_name*, respecting per-agent overrides.
-    return get_llm(model=settings.LLM_ROUTER_MODEL, temperature=temperature)
+
+    Falls back to ``settings.LLM_MODEL`` for unknown agent names or when the
+    per-agent override is left empty in ``.env``.
+    """
+    model = model_for_agent(agent_name)
+    return get_llm(model=model, temperature=temperature)
 
 
 async def embed(text: str) -> list[float]:

app/core/memory_extraction.py

@@ -0,0 +1,450 @@
+"""Mem0-style Extract/Update pipeline — Phase 2.
+
+Runs after every ``store_episode`` call to distil durable facts, preferences,
+routines, and relations from the latest conversation turn.
+
+Entry point: ``run_extraction(db, user_id, last_user_msg, last_assistant_msg, session_id)``
+
+Design notes
+------------
+- Two gpt-4o-mini calls per turn: extract candidates, then decide action per candidate.
+- Short-circuit: if no existing neighbours → ADD without a second LLM call (cost saving).
+- Zero-trust: never logs decrypted user content; relation subject/object labels are
+  treated as identifiers (safe to log per spec).
+- Must not raise into the request path — caller wraps in asyncio.create_task().
+"""
+
+from __future__ import annotations
+
+import json
+import logging
+from typing import Any, Literal
+
+from pydantic import BaseModel, Field
+from sqlalchemy.ext.asyncio import AsyncSession
+
+from app.core.langfuse_client import get_langfuse, get_prompt_or_fallback, extract_usage, langfuse_context
+from app.core.llm import get_agent_llm, model_for_agent
+
+logger = logging.getLogger(__name__)
+
+# ── Fallback prompts (used when Langfuse unavailable) ─────────────────────────
+
+_EXTRACTION_FALLBACK = (
+    "You are a memory extractor for a personal AI secretary. Given the last conversation "
+    "turn, the user's core memory, and recent episode summaries, identify durable facts, "
+    "preferences, routines, and person/project relations worth remembering.\n\n"
+    "Output JSON matching this schema exactly:\n"
+    '{{"candidates": [{{"type": "<fact|preference|relation|routine>", '
+    '"content": "<short canonical statement>", '
+    '"target_tier": "<core|associative|relational|proactive>", '
+    '"subject": null, "predicate": null, "object": null, "confidence": 0.7}}]}}\n\n'
+    "Rules:\n"
+    "- Skip small talk, greetings, one-off questions.\n"
+    "- Max 5 candidates per call.\n"
+    "- Only extract durable information (still true next week).\n"
+    "- For type=relation: subject/predicate/object required.\n"
+    "- Default confidence=0.7.\n\n"
+    "## Last turn\n{last_turn}\n\n"
+    "## Core memory (current)\n{core_memory}\n\n"
+    "## Recent episodes\n{recent_episodes}"
+)
+
+_DECIDE_FALLBACK = (
+    "You are a memory update decision engine. Given a new memory candidate and a list of "
+    "existing memories from the same tier, decide what action to take.\n\n"
+    "Respond with exactly one word: ADD, UPDATE, DELETE, or NOOP.\n\n"
+    "- ADD: new information not in existing memories.\n"
+    "- UPDATE: contradicts or supersedes an existing memory.\n"
+    "- DELETE: states something is no longer true.\n"
+    "- NOOP: already captured accurately.\n\n"
+    "## New candidate\n{candidate}\n\n"
+    "## Existing memories (same tier, top neighbours)\n{existing_memories}"
+)
+
+
+# ── Pydantic schemas ───────────────────────────────────────────────────────────
+
+class MemoryCandidate(BaseModel):
+    type: Literal["fact", "preference", "relation", "routine"]
+    content: str
+    target_tier: Literal["core", "associative", "relational", "proactive"]
+    subject: str | None = None
+    predicate: str | None = None
+    object: str | None = None
+    confidence: float = Field(default=0.7, ge=0.0, le=1.0)
+
+
+class ExtractionResult(BaseModel):
+    candidates: list[MemoryCandidate] = Field(default_factory=list)
+
+
+# ── Task 2.1 — Extract candidates ─────────────────────────────────────────────
+
+async def extract_candidates(
+    last_turn: str,
+    core_memory: dict[str, str],
+    recent_episodes: list[str],
+) -> ExtractionResult:
+    """Call gpt-4o-mini to extract memory candidates from the latest turn.
+
+    Returns an ExtractionResult (may be empty on failure — never raises).
+    """
+    core_str = "\n".join(f"{k}: {v}" for k, v in core_memory.items()) or "(empty)"
+    episodes_str = "\n---\n".join(recent_episodes[-5:]) or "(none)"
+
+    template, prompt_obj = get_prompt_or_fallback("memory_extraction", _EXTRACTION_FALLBACK)
+
+    # Compile with Langfuse variable syntax ({{var}}) or fallback {var}
+    if prompt_obj is not None:
+        try:
+            system_text = prompt_obj.compile(
+                last_turn=last_turn,
+                core_memory=core_str,
+                recent_episodes=episodes_str,
+            )
+            if isinstance(system_text, list):
+                system_text = "\n".join(m.get("content", "") for m in system_text if isinstance(m, dict))
+        except Exception as exc:
+            logger.warning("memory_extraction: compile failed: %s", exc)
+            system_text = template.format(
+                last_turn=last_turn,
+                core_memory=core_str,
+                recent_episodes=episodes_str,
+            )
+    else:
+        system_text = template.format(
+            last_turn=last_turn,
+            core_memory=core_str,
+            recent_episodes=episodes_str,
+        )
+
+    llm = get_agent_llm("memory-extractor", temperature=0)
+    # Bind JSON mode so the model always returns parseable output.
+    llm_json = llm.bind(response_format={"type": "json_object"})  # type: ignore[attr-defined]
+
+    lf = get_langfuse()
+    try:
+        from langchain_core.messages import HumanMessage, SystemMessage  # noqa: PLC0415
+        messages = [
+            SystemMessage(content=system_text),
+            HumanMessage(content="Extract memory candidates as JSON."),
+        ]
+
+        if lf:
+            with lf.start_as_current_observation(
+                as_type="generation",
+                name="memory-extraction",
+                model=model_for_agent("memory-extractor"),
+                prompt=prompt_obj,
+                input=messages,
+            ) as gen:
+                response = await llm_json.ainvoke(messages)
+                gen.update(output=response.content, usage=extract_usage(response))
+        else:
+            response = await llm_json.ainvoke(messages)
+
+        raw = json.loads(response.content)
+        result = ExtractionResult.model_validate(raw)
+        logger.info("memory_extraction: extracted %d candidates", len(result.candidates))
+        return result
+
+    except Exception as exc:
+        logger.warning("memory_extraction: extract_candidates failed: %s", exc)
+        return ExtractionResult(candidates=[])
+
+
+# ── Task 2.2 — Decide action ──────────────────────────────────────────────────
+
+async def decide_action(
+    candidate: MemoryCandidate,
+    existing: list[str],
+) -> Literal["ADD", "UPDATE", "DELETE", "NOOP"]:
+    """Decide what to do with a candidate given existing memories in the same tier.
+
+    Short-circuits to ADD without an LLM call when existing is empty (cost saving).
+    Never raises.
+    """
+    if not existing:
+        return "ADD"
+
+    candidate_str = f"[{candidate.type}] {candidate.content}"
+    existing_str = "\n".join(f"- {m}" for m in existing)
+
+    template, prompt_obj = get_prompt_or_fallback("memory_decide_action", _DECIDE_FALLBACK)
+
+    if prompt_obj is not None:
+        try:
+            system_text = prompt_obj.compile(
+                candidate=candidate_str,
+                existing_memories=existing_str,
+            )
+            if isinstance(system_text, list):
+                system_text = "\n".join(m.get("content", "") for m in system_text if isinstance(m, dict))
+        except Exception as exc:
+            logger.warning("memory_extraction: decide compile failed: %s", exc)
+            system_text = template.format(candidate=candidate_str, existing_memories=existing_str)
+    else:
+        system_text = template.format(candidate=candidate_str, existing_memories=existing_str)
+
+    llm = get_agent_llm("memory-extractor", temperature=0)
+    lf = get_langfuse()
+
+    try:
+        from langchain_core.messages import HumanMessage, SystemMessage  # noqa: PLC0415
+        messages = [
+            SystemMessage(content=system_text),
+            HumanMessage(content="Decide action."),
+        ]
+
+        if lf:
+            with lf.start_as_current_observation(
+                as_type="generation",
+                name="memory-decide-action",
+                model=model_for_agent("memory-extractor"),
+                prompt=prompt_obj,
+                input=messages,
+            ) as gen:
+                response = await llm.ainvoke(messages)
+                gen.update(output=response.content, usage=extract_usage(response))
+        else:
+            response = await llm.ainvoke(messages)
+
+        verb = response.content.strip().upper()
+        if verb in ("ADD", "UPDATE", "DELETE", "NOOP"):
+            return verb  # type: ignore[return-value]
+        logger.warning("memory_extraction: unexpected decide verb=%r, defaulting ADD", verb)
+        return "ADD"
+
+    except Exception as exc:
+        logger.warning("memory_extraction: decide_action failed: %s", exc)
+        return "ADD"
+
+
+# ── Task 2.3 — Pipeline orchestrator ──────────────────────────────────────────
+
+async def run_extraction(
+    db: AsyncSession,
+    user_id: str,
+    last_user_msg: str,
+    last_assistant_msg: str,
+    session_id: str | None,
+) -> None:
+    """Full Mem0-style extract/update pipeline for one conversation turn.
+
+    Steps:
+    1. Load core memory + last 5 episodes.
+    2. extract_candidates() → up to 5 MemoryCandidate objects.
+    3. For each candidate: find top-3 neighbours → decide_action() → apply.
+    4. Trace via Langfuse.
+
+    Never raises — wraps everything in try/except.
+    """
+    try:
+        await _run_extraction_inner(db, user_id, last_user_msg, last_assistant_msg, session_id)
+    except Exception as exc:
+        logger.warning("memory_extraction: run_extraction failed user=%s: %s", user_id, exc)
+
+
+async def _run_extraction_inner(
+    db: AsyncSession,
+    user_id: str,
+    last_user_msg: str,
+    last_assistant_msg: str,
+    session_id: str | None,
+) -> None:
+    from app.core.memory_middleware import MemoryMiddleware  # noqa: PLC0415
+
+    middleware = MemoryMiddleware(db)
+    fernet = await middleware._get_fernet(user_id)
+    if fernet is None:
+        logger.warning("memory_extraction: no fernet for user=%s, skipping", user_id)
+        return
+
+    # 1. Load context
+    core: dict[str, str] = await middleware._load_core(user_id, fernet)
+    episodes: list[str] = await middleware._load_episodic(user_id, fernet, session_id=session_id)
+
+    last_turn = f"User: {last_user_msg}\nAssistant: {last_assistant_msg}"
+
+    lf = get_langfuse()
+
+    async def _run(trace_id: str | None) -> dict[str, Any]:
+        # 2. Extract candidates
+        result = await extract_candidates(last_turn, core, episodes)
+        if not result.candidates:
+            logger.info("memory_extraction: no candidates user=%s", user_id)
+            return {"candidates": 0, "applied": 0}
+
+        logger.info(
+            "memory_extraction: processing %d candidates user=%s trace=%s",
+            len(result.candidates),
+            user_id,
+            trace_id or "-",
+        )
+
+        # 3. Apply each candidate
+        applied = 0
+        actions: list[str] = []
+        for candidate in result.candidates:
+            try:
+                await _apply_candidate(middleware, db, user_id, fernet, candidate, trace_id)
+                applied += 1
+                actions.append(f"{candidate.type}:{candidate.target_tier}")
+            except Exception as exc:
+                logger.warning(
+                    "memory_extraction: apply failed candidate=%r user=%s: %s",
+                    candidate.content[:80],
+                    user_id,
+                    exc,
+                )
+
+        logger.info(
+            "memory_extraction: applied %d/%d candidates user=%s",
+            applied,
+            len(result.candidates),
+            user_id,
+        )
+        return {"candidates": len(result.candidates), "applied": applied, "actions": actions}
+
+    with langfuse_context(user_id=user_id, session_id=session_id):
+        if lf:
+            with lf.start_as_current_observation(
+                as_type="span",
+                name="memory-extraction-pipeline",
+                input={"last_turn_preview": last_turn[:200]},
+            ) as span:
+                summary = await _run(trace_id=span.id)
+                span.update(output=summary)
+            try:
+                lf.flush()
+            except Exception:
+                pass
+        else:
+            await _run(trace_id=None)
+
+
+async def _apply_candidate(
+    middleware: Any,
+    db: AsyncSession,
+    user_id: str,
+    fernet: Any,
+    candidate: MemoryCandidate,
+    trace_id: str | None,
+) -> None:
+    """Fetch neighbours, decide action, apply to the appropriate tier."""
+
+    neighbours: list[str] = []
+
+    if candidate.target_tier == "core":
+        # For core tier: neighbours are existing core block values for similar keys.
+        blocks = await middleware.list_core_blocks(user_id)
+        neighbours = [b["value"] for b in blocks[:3]]
+
+    elif candidate.target_tier == "associative":
+        neighbours = await middleware.search_archival(user_id, candidate.content, top_k=3)
+
+    elif candidate.target_tier == "relational":
+        # Relation candidates handled specially — passed to upsert_relation directly.
+        # Neighbours: search by subject label if available.
+        neighbours = []
+
+    elif candidate.target_tier == "proactive":
+        neighbours = await middleware.search_recall(user_id, candidate.content, top_k=3)
+
+    action = await decide_action(candidate, neighbours)
+    logger.info(
+        "memory_extraction: candidate type=%s tier=%s action=%s",
+        candidate.type,
+        candidate.target_tier,
+        action,
+    )
+
+    if action == "NOOP":
+        return
+
+    if candidate.target_tier == "relational":
+        # Always upsert relations — decide_action skipped (no neighbour search).
+        if candidate.subject and candidate.predicate and candidate.object:
+            await _upsert_relation(
+                middleware, db, user_id, candidate, trace_id
+            )
+        return
+
+    if action in ("ADD", "UPDATE"):
+        if candidate.target_tier == "core":
+            # Derive a short key from the content (first 40 chars, snake_cased).
+            key = _content_to_key(candidate.content)
+            await middleware.update_core(user_id, key, candidate.content, trace_id=trace_id)
+
+        elif candidate.target_tier == "associative":
+            await middleware.store_associative(user_id, candidate.content)
+
+        elif candidate.target_tier == "proactive":
+            await _store_proactive_stub(middleware, db, user_id, candidate, fernet)
+
+    elif action == "DELETE":
+        if candidate.target_tier == "core":
+            key = _content_to_key(candidate.content)
+            await middleware.delete_core(user_id, key)
+
+
+def _content_to_key(content: str) -> str:
+    """Derive a short snake_case key from a content string (first 40 chars)."""
+    import re  # noqa: PLC0415
+    slug = re.sub(r"[^a-z0-9]+", "_", content[:40].lower()).strip("_")
+    return slug or "memory"
+
+
+async def _upsert_relation(
+    middleware: Any,
+    db: AsyncSession,
+    user_id: str,
+    candidate: MemoryCandidate,
+    trace_id: str | None,
+) -> None:
+    """Upsert a relation row via MemoryMiddleware.upsert_relation (Phase 3)."""
+    await middleware.upsert_relation(
+        user_id=user_id,
+        subject=candidate.subject or "unknown",
+        subject_type="unknown",
+        predicate=candidate.predicate or "related_to",
+        object_=candidate.object or "unknown",
+        object_type="unknown",
+        confidence=candidate.confidence,
+    )
+    logger.info(
+        "memory_extraction: upserted relation subject=%s predicate=%s object=%s",
+        candidate.subject,
+        candidate.predicate,
+        candidate.object,
+    )
+
+
+async def _store_proactive_stub(
+    middleware: Any,
+    db: AsyncSession,
+    user_id: str,
+    candidate: MemoryCandidate,
+    fernet: Any,
+) -> None:
+    """Store a proactive pattern row directly (MemoryProactive model)."""
+    import uuid  # noqa: PLC0415
+    from app.models import MemoryProactive  # noqa: PLC0415
+    from app.core.memory_middleware import _encrypt  # noqa: PLC0415
+
+    encrypted = _encrypt(fernet, candidate.content)
+    row = MemoryProactive(
+        id=str(uuid.uuid4()),
+        user_id=user_id,
+        pattern_encrypted=encrypted,
+        confidence=candidate.confidence,
+        source="inferred",
+    )
+    db.add(row)
+    try:
+        await db.commit()
+        logger.info("memory_extraction: stored proactive pattern user=%s", user_id)
+    except Exception as exc:
+        logger.warning("memory_extraction: store proactive failed: %s", exc)
+        await db.rollback()

app/core/memory_maintenance.py

@@ -0,0 +1,581 @@
+"""Memory maintenance jobs — Phase 3/5.
+
+Three entrypoints called by the scheduler (APScheduler) registered in app/main.py:
+
+  drain_extraction_queue(db) — Free-tier batch extraction (Phase 2/5).
+  mine_proactive_patterns(db, user_id) — Power+ pattern mining (Phase 5).
+  decay_relations(db, user_id) — confidence decay + pruning for memory_relations (Phase 3).
+
+All are safe to call manually or from tests; they never raise.
+"""
+
+from __future__ import annotations
+
+import json
+import logging
+import uuid
+from datetime import datetime, timedelta, timezone
+
+from cryptography.fernet import Fernet
+from sqlalchemy import select
+from sqlalchemy.ext.asyncio import AsyncSession
+
+from app.core.langfuse_client import compile_prompt, extract_usage, get_langfuse, get_prompt_or_fallback
+from app.models import MemoryAssociative, MemoryEpisodic, MemoryProactive, MemoryRelation, User
+
+logger = logging.getLogger(__name__)
+
+# Decay parameters for relations
+_DECAY_FACTOR = 0.95
+_DECAY_PERIOD_DAYS = 30
+_PRUNE_THRESHOLD = 0.2
+
+# Proactive pattern decay: 10 % per 7 days since last sighting
+_PROACTIVE_DECAY_FACTOR = 0.9
+_PROACTIVE_DECAY_PERIOD_DAYS = 7
+_PROACTIVE_PRUNE_THRESHOLD = 0.2
+
+# Mining: require at least this many episodes to attempt pattern extraction
+_MIN_EPISODES_FOR_MINING = 3
+_MINING_LOOKBACK_DAYS = 30
+
+# Audit: caps to control token cost
+_AUDIT_MAX_FACTS = 50
+_AUDIT_MAX_LABELS = 100
+
+
+async def decay_relations(db: AsyncSession, user_id: str) -> None:
+    """Apply confidence decay to all relation rows for a user.
+
+    Decay rule: confidence *= 0.95 for every 30 days since last_confirmed_at.
+    Rows whose confidence falls below 0.2 are deleted.
+
+    Never raises — wraps in try/except.
+    """
+    try:
+        await _decay_relations_inner(db, user_id)
+    except Exception as exc:
+        logger.warning("memory_maintenance: decay_relations failed user=%s: %s", user_id, exc)
+
+
+async def _decay_relations_inner(db: AsyncSession, user_id: str) -> None:
+    result = await db.execute(
+        select(MemoryRelation).where(MemoryRelation.user_id == user_id)
+    )
+    rows = result.scalars().all()
+    now = datetime.now(timezone.utc)
+    deleted = 0
+    decayed = 0
+
+    for row in rows:
+        reference = row.last_confirmed_at or row.created_at
+        if reference is None:
+            continue
+        if reference.tzinfo is None:
+            reference = reference.replace(tzinfo=timezone.utc)
+
+        days_elapsed = (now - reference).days
+        if days_elapsed < _DECAY_PERIOD_DAYS:
+            continue
+
+        periods = days_elapsed // _DECAY_PERIOD_DAYS
+        new_confidence = row.confidence * (_DECAY_FACTOR ** periods)
+
+        if new_confidence < _PRUNE_THRESHOLD:
+            await db.delete(row)
+            deleted += 1
+            logger.info(
+                "memory_maintenance: pruned relation id=%s user=%s subject=%s predicate=%s "
+                "confidence=%.3f (below threshold)",
+                row.id, user_id, row.subject_label, row.predicate, new_confidence,
+            )
+        else:
+            row.confidence = new_confidence
+            decayed += 1
+
+    try:
+        await db.commit()
+        logger.info(
+            "memory_maintenance: decay_relations user=%s decayed=%d deleted=%d",
+            user_id, decayed, deleted,
+        )
+    except Exception as exc:
+        logger.warning("memory_maintenance: decay_relations commit failed user=%s: %s", user_id, exc)
+        await db.rollback()
+
+
+async def drain_extraction_queue(db: AsyncSession) -> None:
+    """Process pending ExtractionQueue rows for Free-tier users.
+
+    Each row corresponds to a stored episode that should be fed through the
+    Mem0-style extraction pipeline. Rows are deleted after successful processing.
+    Never raises — wraps in try/except.
+    """
+    try:
+        await _drain_extraction_queue_inner(db)
+    except Exception as exc:
+        logger.warning("memory_maintenance: drain_extraction_queue failed: %s", exc)
+
+
+async def _drain_extraction_queue_inner(db: AsyncSession) -> None:
+    from app.models import ExtractionQueue  # noqa: PLC0415
+
+    result = await db.execute(select(ExtractionQueue))
+    rows = result.scalars().all()
+
+    if not rows:
+        logger.debug("memory_maintenance: drain_extraction_queue nothing to drain")
+        return
+
+    logger.info("memory_maintenance: drain_extraction_queue pending=%d", len(rows))
+
+    from app.core.memory_extraction import run_extraction  # noqa: PLC0415
+
+    processed = 0
+    for row in rows:
+        try:
+            await run_extraction(
+                db=db,
+                user_id=row.user_id,
+                last_user_msg="",
+                last_assistant_msg="",
+                session_id=None,
+            )
+            await db.delete(row)
+            await db.commit()
+            processed += 1
+        except Exception as exc:
+            logger.warning(
+                "memory_maintenance: drain failed row=%s user=%s: %s",
+                row.id, row.user_id, exc,
+            )
+            await db.rollback()
+
+    logger.info("memory_maintenance: drain_extraction_queue processed=%d/%d", processed, len(rows))
+
+
+async def mine_proactive_patterns(db: AsyncSession, user_id: str) -> None:
+    """Mine recurring behavioral patterns from last 30 days of episodes (Power+ only).
+
+    Steps:
+    1. Gate on proactive_mining tier feature.
+    2. Load + decrypt last 30 days of episodic summaries.
+    3. Call gpt-4o-mini to identify recurring patterns.
+    4. Encrypt and store each pattern in memory_proactive.
+    5. Apply decay to existing proactive rows.
+
+    Never raises — wraps in try/except.
+    """
+    try:
+        await _mine_proactive_patterns_inner(db, user_id)
+    except Exception as exc:
+        logger.warning("memory_maintenance: mine_proactive_patterns failed user=%s: %s", user_id, exc)
+
+
+async def _mine_proactive_patterns_inner(db: AsyncSession, user_id: str) -> None:
+    from app.billing.tier_manager import tier_manager  # noqa: PLC0415
+
+    tier = await tier_manager.get_tier(user_id, db)
+    if not tier_manager.check_feature(tier, "proactive_mining"):
+        logger.debug("memory_maintenance: mine_proactive_patterns skipped (tier=%s)", tier)
+        return
+
+    # Load user Fernet key
+    result = await db.execute(select(User).where(User.id == user_id))
+    user = result.scalar_one_or_none()
+    if user is None or not user.encryption_key:
+        logger.warning("memory_maintenance: mine_proactive_patterns no encryption_key user=%s", user_id)
+        return
+
+    fernet = Fernet(user.encryption_key.encode())
+    cutoff = datetime.now(timezone.utc) - timedelta(days=_MINING_LOOKBACK_DAYS)
+
+    episodes_result = await db.execute(
+        select(MemoryEpisodic)
+        .where(
+            MemoryEpisodic.user_id == user_id,
+            MemoryEpisodic.created_at >= cutoff,
+        )
+        .order_by(MemoryEpisodic.created_at.asc())
+    )
+    episode_rows = episodes_result.scalars().all()
+
+    if len(episode_rows) < _MIN_EPISODES_FOR_MINING:
+        logger.info(
+            "memory_maintenance: mine_proactive_patterns skipped user=%s episodes=%d (< %d)",
+            user_id, len(episode_rows), _MIN_EPISODES_FOR_MINING,
+        )
+        return
+
+    summaries: list[str] = []
+    for ep in episode_rows:
+        try:
+            plaintext = fernet.decrypt(ep.summary_encrypted.encode()).decode()
+            summaries.append(plaintext)
+        except Exception:
+            pass
+
+    if not summaries:
+        return
+
+    patterns = await _extract_proactive_patterns(summaries)
+    if not patterns:
+        logger.info("memory_maintenance: mine_proactive_patterns user=%s no patterns extracted", user_id)
+        return
+
+    stored = 0
+    for pattern_text in patterns:
+        try:
+            encrypted = fernet.encrypt(pattern_text.encode()).decode()
+            row = MemoryProactive(
+                id=str(uuid.uuid4()),
+                user_id=user_id,
+                pattern_encrypted=encrypted,
+                confidence=0.7,
+                source="inferred",
+            )
+            db.add(row)
+            stored += 1
+        except Exception as exc:
+            logger.warning("memory_maintenance: failed to store pattern user=%s: %s", user_id, exc)
+
+    try:
+        await db.commit()
+        logger.info(
+            "memory_maintenance: mine_proactive_patterns user=%s stored=%d",
+            user_id, stored,
+        )
+    except Exception as exc:
+        logger.warning("memory_maintenance: mine_proactive_patterns commit failed user=%s: %s", user_id, exc)
+        await db.rollback()
+        return
+
+    await _decay_proactive_patterns(db, user_id, fernet)
+
+
+async def _extract_proactive_patterns(summaries: list[str]) -> list[str]:
+    """Call memory-miner LLM to identify recurring behavioral/temporal patterns."""
+    from app.core.llm import get_agent_llm  # noqa: PLC0415
+
+    llm = get_agent_llm("memory-miner", temperature=0)
+    combined = "\n---\n".join(summaries[-20:])  # cap at last 20 to control token usage
+    prompt = (
+        "You are analyzing conversation history for a personal AI secretary. "
+        "Identify 3-5 recurring temporal or behavioral patterns (e.g. 'always works late on Thursdays', "
+        "'prefers bullet-point summaries', 'frequently asks about Project Acme status'). "
+        "Return each pattern as a plain, short English sentence on its own line. "
+        "No numbering, no bullet points, no extra text.\n\n"
+        f"Conversation history:\n{combined}"
+    )
+    try:
+        response = await llm.ainvoke(prompt)
+        text = response.content if hasattr(response, "content") else str(response)
+        lines = [line.strip() for line in str(text).splitlines() if line.strip()]
+        return lines[:5]
+    except Exception as exc:
+        logger.warning("memory_maintenance: _extract_proactive_patterns LLM failed: %s", exc)
+        return []
+
+
+async def _decay_proactive_patterns(db: AsyncSession, user_id: str, fernet: Fernet) -> None:
+    """Decay confidence of existing proactive patterns; prune below threshold."""
+    result = await db.execute(
+        select(MemoryProactive).where(MemoryProactive.user_id == user_id)
+    )
+    rows = result.scalars().all()
+    now = datetime.now(timezone.utc)
+    deleted = 0
+    decayed = 0
+
+    for row in rows:
+        reference = row.created_at
+        if reference is None:
+            continue
+        if reference.tzinfo is None:
+            reference = reference.replace(tzinfo=timezone.utc)
+
+        days_elapsed = (now - reference).days
+        if days_elapsed < _PROACTIVE_DECAY_PERIOD_DAYS:
+            continue
+
+        periods = days_elapsed // _PROACTIVE_DECAY_PERIOD_DAYS
+        new_confidence = row.confidence * (_PROACTIVE_DECAY_FACTOR ** periods)
+
+        if new_confidence < _PROACTIVE_PRUNE_THRESHOLD:
+            await db.delete(row)
+            deleted += 1
+        else:
+            row.confidence = new_confidence
+            decayed += 1
+
+    try:
+        await db.commit()
+        logger.info(
+            "memory_maintenance: decay_proactive user=%s decayed=%d deleted=%d",
+            user_id, decayed, deleted,
+        )
+    except Exception as exc:
+        logger.warning("memory_maintenance: decay_proactive commit failed user=%s: %s", user_id, exc)
+        await db.rollback()
+
+
+# ── Phase 7: weekly memory audit ──────────────────────────────────────────────
+
+_AUDIT_CONTRADICTIONS_FALLBACK = (
+    "You are auditing a personal AI assistant's memory bank. "
+    "Each fact has an ID in brackets. "
+    "Find pairs that directly contradict each other "
+    "(e.g. 'prefers morning meetings' vs 'never schedules before noon'). "
+    "For each contradiction, pick the ID to DELETE (the older or less specific one). "
+    'Return ONLY a valid JSON array, no markdown fences: '
+    '[{{"delete": "<id>", "reason": "<one line>"}}]. '
+    "If no contradictions, return [].\n\n"
+    "Facts:\n{facts}"
+)
+
+_AUDIT_CANONICALIZE_FALLBACK = (
+    "You are auditing entity labels in a personal AI assistant's relational memory. "
+    "These are names of people, companies, projects, or topics. "
+    "Group labels that clearly refer to the same real-world entity "
+    "(e.g. 'giulia', 'Giulia', 'Giulia R.' → canonical 'Giulia'). "
+    "Return ONLY a valid JSON array, no markdown fences: "
+    '[{{"canonical": "<best label>", "variants": ["<v1>", "<v2>"]}}]. '
+    "Only include groups with at least one variant. Singletons: omit.\n\n"
+    "Labels:\n{labels}"
+)
+
+
+async def audit_memory(db: AsyncSession, user_id: str) -> None:
+    """Weekly audit: contradiction scan on associative facts + label canonicalization on relations.
+
+    Steps:
+    1. Decrypt up to _AUDIT_MAX_FACTS associative rows; send list to memory-auditor LLM.
+    2. LLM flags rows to delete (direct contradictions); hard-delete them.
+    3. Collect unique subject/object labels from memory_relations; ask LLM to group duplicates.
+    4. Rewrite variant labels to their canonical form in-place.
+
+    Never raises — wraps in try/except.
+    """
+    try:
+        await _audit_memory_inner(db, user_id)
+    except Exception as exc:
+        logger.warning("memory_maintenance: audit_memory failed user=%s: %s", user_id, exc)
+
+
+async def _audit_memory_inner(db: AsyncSession, user_id: str) -> None:
+    result = await db.execute(select(User).where(User.id == user_id))
+    user = result.scalar_one_or_none()
+    if user is None or not user.encryption_key:
+        logger.warning("memory_maintenance: audit_memory no encryption_key user=%s", user_id)
+        return
+
+    fernet = Fernet(user.encryption_key.encode())
+    await _scan_associative_contradictions(db, user_id, fernet)
+    await _canonicalize_relation_labels(db, user_id)
+
+
+async def _scan_associative_contradictions(
+    db: AsyncSession,
+    user_id: str,
+    fernet: Fernet,
+) -> None:
+    """Decrypt associative facts, ask LLM to flag contradictions, delete superseded rows."""
+    result = await db.execute(
+        select(MemoryAssociative)
+        .where(MemoryAssociative.user_id == user_id)
+        .order_by(MemoryAssociative.updated_at.desc())
+        .limit(_AUDIT_MAX_FACTS)
+    )
+    rows = result.scalars().all()
+    if len(rows) < 2:
+        return
+
+    id_to_text: dict[str, str] = {}
+    for row in rows:
+        try:
+            plaintext = fernet.decrypt(row.content_encrypted.encode()).decode()
+            id_to_text[row.id] = plaintext
+        except Exception:
+            pass
+
+    if len(id_to_text) < 2:
+        return
+
+    id_list = list(id_to_text.keys())
+    numbered = "\n".join(
+        f"{i + 1}. [{rid}] {id_to_text[rid]}" for i, rid in enumerate(id_list)
+    )
+
+    template, prompt_obj = get_prompt_or_fallback(
+        "memory_audit_contradictions", _AUDIT_CONTRADICTIONS_FALLBACK
+    )
+    system_text = compile_prompt(template, prompt_obj, facts=numbered)
+
+    from app.core.llm import get_agent_llm, model_for_agent  # noqa: PLC0415
+    from langchain_core.messages import HumanMessage, SystemMessage  # noqa: PLC0415
+
+    llm = get_agent_llm("memory-auditor", temperature=0)
+    lf = get_langfuse()
+    messages = [
+        SystemMessage(content=system_text),
+        HumanMessage(content="Audit facts for contradictions."),
+    ]
+    try:
+        if lf:
+            with lf.start_as_current_observation(
+                as_type="generation",
+                name="memory-audit-contradictions",
+                model=model_for_agent("memory-auditor"),
+                prompt=prompt_obj,
+                input=messages,
+            ) as gen:
+                response = await llm.ainvoke(messages)
+                gen.update(output=response.content, usage=extract_usage(response))
+        else:
+            response = await llm.ainvoke(messages)
+
+        text = response.content if hasattr(response, "content") else str(response)
+        deletions = json.loads(text.strip())
+        if not isinstance(deletions, list):
+            return
+    except Exception as exc:
+        logger.warning(
+            "memory_maintenance: _scan_associative_contradictions LLM/parse failed user=%s: %s",
+            user_id, exc,
+        )
+        return
+
+    deleted = 0
+    for item in deletions:
+        if not isinstance(item, dict):
+            continue
+        rid = item.get("delete")
+        if not rid or rid not in id_to_text:
+            continue
+        result2 = await db.execute(
+            select(MemoryAssociative).where(
+                MemoryAssociative.id == rid,
+                MemoryAssociative.user_id == user_id,
+            )
+        )
+        target = result2.scalar_one_or_none()
+        if target:
+            await db.delete(target)
+            deleted += 1
+            logger.info(
+                "memory_maintenance: audit deleted contradiction id=%s user=%s reason=%s",
+                rid, user_id, item.get("reason", ""),
+            )
+
+    if deleted:
+        try:
+            await db.commit()
+        except Exception as exc:
+            logger.warning(
+                "memory_maintenance: audit contradiction commit failed user=%s: %s", user_id, exc
+            )
+            await db.rollback()
+
+    logger.info(
+        "memory_maintenance: _scan_associative_contradictions user=%s deleted=%d", user_id, deleted
+    )
+
+
+async def _canonicalize_relation_labels(db: AsyncSession, user_id: str) -> None:
+    """Group near-duplicate entity labels in memory_relations and unify to canonical form."""
+    result = await db.execute(
+        select(MemoryRelation).where(MemoryRelation.user_id == user_id)
+    )
+    rows = result.scalars().all()
+    if not rows:
+        return
+
+    all_labels: set[str] = set()
+    for row in rows:
+        all_labels.add(row.subject_label)
+        all_labels.add(row.object_label)
+
+    labels_list = sorted(all_labels)[:_AUDIT_MAX_LABELS]
+    if len(labels_list) < 2:
+        return
+
+    labels_block = "\n".join(f"- {lbl}" for lbl in labels_list)
+    template, prompt_obj = get_prompt_or_fallback(
+        "memory_audit_canonicalize", _AUDIT_CANONICALIZE_FALLBACK
+    )
+    system_text = compile_prompt(template, prompt_obj, labels=labels_block)
+
+    from app.core.llm import get_agent_llm, model_for_agent  # noqa: PLC0415
+    from langchain_core.messages import HumanMessage, SystemMessage  # noqa: PLC0415
+
+    llm = get_agent_llm("memory-auditor", temperature=0)
+    lf = get_langfuse()
+    messages = [
+        SystemMessage(content=system_text),
+        HumanMessage(content="Canonicalize entity labels."),
+    ]
+    try:
+        if lf:
+            with lf.start_as_current_observation(
+                as_type="generation",
+                name="memory-audit-canonicalize",
+                model=model_for_agent("memory-auditor"),
+                prompt=prompt_obj,
+                input=messages,
+            ) as gen:
+                response = await llm.ainvoke(messages)
+                gen.update(output=response.content, usage=extract_usage(response))
+        else:
+            response = await llm.ainvoke(messages)
+
+        text = response.content if hasattr(response, "content") else str(response)
+        groups = json.loads(text.strip())
+        if not isinstance(groups, list):
+            return
+    except Exception as exc:
+        logger.warning(
+            "memory_maintenance: _canonicalize_relation_labels LLM/parse failed user=%s: %s",
+            user_id, exc,
+        )
+        return
+
+    # Build variant → canonical map
+    remap: dict[str, str] = {}
+    for group in groups:
+        if not isinstance(group, dict):
+            continue
+        canonical = group.get("canonical", "")
+        variants = group.get("variants") or []
+        if not canonical:
+            continue
+        for v in variants:
+            if isinstance(v, str) and v != canonical:
+                remap[v] = canonical
+
+    if not remap:
+        return
+
+    updated = 0
+    for row in rows:
+        changed = False
+        if row.subject_label in remap:
+            row.subject_label = remap[row.subject_label]
+            changed = True
+        if row.object_label in remap:
+            row.object_label = remap[row.object_label]
+            changed = True
+        if changed:
+            updated += 1
+
+    if updated:
+        try:
+            await db.commit()
+            logger.info(
+                "memory_maintenance: _canonicalize_relation_labels user=%s updated=%d",
+                user_id, updated,
+            )
+        except Exception as exc:
+            logger.warning(
+                "memory_maintenance: canonicalize commit failed user=%s: %s", user_id, exc
+            )
+            await db.rollback()

app/core/memory_middleware.py

@@ -18,8 +18,10 @@ Usage:
 
 from __future__ import annotations
 
+import asyncio
 import logging
 import uuid
+from datetime import datetime, timezone
 from typing import Any
 
 from cryptography.fernet import Fernet, InvalidToken
@@ -27,15 +29,22 @@ from sqlalchemy import select
 from sqlalchemy.ext.asyncio import AsyncSession
 
 from app.models import (
+    ExtractionQueue,
     MemoryAssociative,
     MemoryCore,
     MemoryEpisodic,
     MemoryProactive,
+    MemoryRelation,
     User,
 )
 
 logger = logging.getLogger(__name__)
 
+
+def _now() -> datetime:
+    return datetime.now(timezone.utc)
+
+
 # Tuning constants
 _ASSOCIATIVE_TOP_K = 5
 _EPISODIC_RECENT_N = 10
@@ -64,26 +73,31 @@ class MemoryMiddleware:
           associative_memory — [plaintext_content, ...]  (top-k by keyword match)
           episodic_memory    — [plaintext_summary, ...]  (most recent N)
           proactive_hints    — [plaintext_pattern, ...]  (above threshold)
+          relational_memory  — ["subject --predicate--> object", ...] (top 10, Pro+)
         """
         fernet = await self._get_fernet(user_id)
         if fernet is None:
             return {}
 
+        user_dbg = await self._get_user_debug(user_id)
+        user_tier: str = user_dbg.get("tier") or "free"
+
         core = await self._load_core(user_id, fernet)
-        associative = await self._load_associative(user_id, message, fernet)
+        associative = await self._load_associative(user_id, message, fernet, user_tier=user_tier)
         episodic = await self._load_episodic(user_id, fernet, session_id=session_id)
         proactive = await self._load_proactive(user_id, fernet)
+        relational = await self._load_relational(user_id, user_tier=user_tier)
 
-        user_dbg = await self._get_user_debug(user_id)
         logger.info(
-            "memory: enrich_context trace=%s user=%s tier=%s core=%d associative=%d episodic=%d proactive=%d",
+            "memory: enrich_context trace=%s user=%s tier=%s core=%d associative=%d episodic=%d proactive=%d relational=%d",
             trace_id or "-",
             user_id,
-            user_dbg.get("tier") or "-",
+            user_tier,
             len(core),
             len(associative),
             len(episodic),
             len(proactive),
+            len(relational),
         )
 
         return {
@@ -91,6 +105,7 @@ class MemoryMiddleware:
             "associative_memory": associative,
             "episodic_memory": episodic,
             "proactive_hints": proactive,
+            "relational_memory": relational,
         }
 
     async def store_episode(
@@ -104,7 +119,10 @@ class MemoryMiddleware:
         """Summarise and store a completed interaction in episodic memory.
 
         The summary is a simple heuristic concatenation (no LLM call) to keep
-        latency low. Full LLM summarisation can be added in a later step.
+        latency low. After committing the episode row, dispatches the Mem0-style
+        extraction pipeline:
+          - Pro/Power/Team → asyncio.create_task (fire-and-forget, realtime).
+          - Free → enqueue an ExtractionQueue row for the daily cron.
         """
         fernet = await self._get_fernet(user_id)
         if fernet is None:
@@ -113,26 +131,95 @@ class MemoryMiddleware:
         summary = f"User: {message[:200]}\nAssistant: {response[:200]}"
         encrypted = _encrypt(fernet, summary)
 
-        row = MemoryEpisodic(
+        episode = MemoryEpisodic(
             id=str(uuid.uuid4()),
             user_id=user_id,
             summary_encrypted=encrypted,
             session_id=session_id,
         )
-        self._db.add(row)
+        self._db.add(episode)
+        episode_id: str = episode.id
         try:
             await self._db.commit()
             user_dbg = await self._get_user_debug(user_id)
+            tier = user_dbg.get("tier") or "free"
             logger.info(
                 "memory: store_episode trace=%s user=%s tier=%s session=%s",
                 trace_id or "-",
                 user_id,
-                user_dbg.get("tier") or "-",
+                tier,
                 session_id,
             )
         except Exception as exc:
             logger.error("memory: store_episode failed user=%s: %s", user_id, exc)
             await self._db.rollback()
+            return
+
+        # ── Dispatch extraction pipeline (Phase 2) ────────────────────────────
+        await self._dispatch_extraction(
+            user_id=user_id,
+            episode_id=episode_id,
+            last_user_msg=message,
+            last_assistant_msg=response,
+            session_id=session_id,
+        )
+
+    async def _dispatch_extraction(
+        self,
+        user_id: str,
+        episode_id: str,
+        last_user_msg: str,
+        last_assistant_msg: str,
+        session_id: str | None,
+    ) -> None:
+        """Route extraction to realtime task or batch queue based on user tier."""
+        from app.billing.tier_manager import tier_manager  # noqa: PLC0415
+
+        tier = await tier_manager.get_tier(user_id, self._db)
+
+        if tier_manager.check_feature(tier, "realtime_extraction"):
+            # Pro/Power/Team: fire-and-forget in the background.
+            # Must open a fresh session — request session closes after handler returns.
+            from app.core.memory_extraction import run_extraction  # noqa: PLC0415
+            from app.db import async_session  # noqa: PLC0415
+
+            async def _task() -> None:
+                try:
+                    async with async_session() as fresh_db:
+                        await run_extraction(
+                            db=fresh_db,
+                            user_id=user_id,
+                            last_user_msg=last_user_msg,
+                            last_assistant_msg=last_assistant_msg,
+                            session_id=session_id,
+                        )
+                except Exception as exc:
+                    logger.warning(
+                        "memory: extraction task failed user=%s: %s", user_id, exc
+                    )
+
+            asyncio.create_task(_task())
+            logger.info("memory: realtime extraction dispatched user=%s", user_id)
+        else:
+            # Free tier: enqueue for daily batch cron.
+            queue_row = ExtractionQueue(
+                id=str(uuid.uuid4()),
+                user_id=user_id,
+                episode_id=episode_id,
+            )
+            self._db.add(queue_row)
+            try:
+                await self._db.commit()
+                logger.info(
+                    "memory: extraction enqueued (batch) user=%s episode=%s",
+                    user_id,
+                    episode_id,
+                )
+            except Exception as exc:
+                logger.warning(
+                    "memory: extraction queue insert failed user=%s: %s", user_id, exc
+                )
+                await self._db.rollback()
 
     async def update_core(self, user_id: str, key: str, value: str, trace_id: str | None = None) -> None:
         """Upsert a core memory key/value for a user."""
@@ -255,6 +342,143 @@ class MemoryMiddleware:
         logger.info("memory: replace_core user=%s label=%s changed=1", user_id, label)
         return True
 
+    async def store_associative(
+        self,
+        user_id: str,
+        content: str,
+        entity_type: str | None = None,
+        entity_id: str | None = None,
+    ) -> None:
+        """Store associative memory; embed if user tier has real_embeddings."""
+        from app.billing.tier_manager import tier_manager  # noqa: PLC0415
+        from app.core.embeddings import embed_text  # noqa: PLC0415
+
+        fernet = await self._get_fernet(user_id)
+        if fernet is None:
+            return
+
+        encrypted = _encrypt(fernet, content)
+
+        user_dbg = await self._get_user_debug(user_id)
+        user_tier = user_dbg.get("tier") or "free"
+
+        embedding: list[float] | None = None
+        if tier_manager.check_feature(user_tier, "real_embeddings"):
+            embedding = await embed_text(content)
+
+        row = MemoryAssociative(
+            id=str(uuid.uuid4()),
+            user_id=user_id,
+            content_encrypted=encrypted,
+            embedding=embedding,
+            entity_type=entity_type,
+            entity_id=entity_id,
+        )
+        self._db.add(row)
+        try:
+            await self._db.commit()
+            logger.info(
+                "memory: store_associative user=%s embedded=%s",
+                user_id,
+                embedding is not None,
+            )
+        except Exception as exc:
+            logger.error("memory: store_associative failed user=%s: %s", user_id, exc)
+            await self._db.rollback()
+
+    async def upsert_relation(
+        self,
+        user_id: str,
+        subject: str,
+        subject_type: str,
+        predicate: str,
+        object_: str,
+        object_type: str,
+        *,
+        confidence: float = 0.7,
+        source_episode_id: str | None = None,
+        notes: str | None = None,
+    ) -> None:
+        """Insert or update a relation row.  Matches on (user_id, subject_label, predicate, object_label).
+
+        subject_label / object_label are plaintext entity identifiers — not encrypted.
+        notes is optional; encrypted with user Fernet if provided.
+        """
+        from app.billing.tier_manager import tier_manager  # noqa: PLC0415
+
+        user_dbg = await self._get_user_debug(user_id)
+        user_tier = user_dbg.get("tier") or "free"
+        if not tier_manager.check_feature(user_tier, "relational_memory"):
+            logger.debug("memory: upsert_relation skipped (tier=%s no relational_memory)", user_tier)
+            return
+
+        notes_encrypted: bytes | None = None
+        if notes:
+            fernet = await self._get_fernet(user_id)
+            if fernet:
+                notes_encrypted = fernet.encrypt(notes.encode())
+
+        result = await self._db.execute(
+            select(MemoryRelation).where(
+                MemoryRelation.user_id == user_id,
+                MemoryRelation.subject_label == subject,
+                MemoryRelation.predicate == predicate,
+                MemoryRelation.object_label == object_,
+            )
+        )
+        existing = result.scalar_one_or_none()
+
+        if existing is not None:
+            existing.subject_type = subject_type
+            existing.object_type = object_type
+            existing.confidence = confidence
+            existing.last_confirmed_at = _now()
+            if notes_encrypted is not None:
+                existing.notes_encrypted = notes_encrypted
+        else:
+            self._db.add(MemoryRelation(
+                id=str(uuid.uuid4()),
+                user_id=user_id,
+                subject_label=subject,
+                subject_type=subject_type,
+                predicate=predicate,
+                object_label=object_,
+                object_type=object_type,
+                confidence=confidence,
+                source_episode_id=source_episode_id,
+                notes_encrypted=notes_encrypted,
+            ))
+
+        try:
+            await self._db.commit()
+            logger.info(
+                "memory: upsert_relation user=%s subject=%s predicate=%s object=%s",
+                user_id, subject, predicate, object_,
+            )
+        except Exception as exc:
+            logger.error("memory: upsert_relation failed user=%s: %s", user_id, exc)
+            await self._db.rollback()
+
+    async def query_relations(
+        self,
+        user_id: str,
+        subject: str | None = None,
+        predicate: str | None = None,
+        object_: str | None = None,
+        limit: int = 20,
+    ) -> list[MemoryRelation]:
+        """Query relation rows for a user with optional filters."""
+        q = select(MemoryRelation).where(MemoryRelation.user_id == user_id)
+        if subject is not None:
+            q = q.where(MemoryRelation.subject_label == subject)
+        if predicate is not None:
+            q = q.where(MemoryRelation.predicate == predicate)
+        if object_ is not None:
+            q = q.where(MemoryRelation.object_label == object_)
+        q = q.order_by(MemoryRelation.confidence.desc()).limit(limit)
+        result = await self._db.execute(q)
+        return list(result.scalars().all())
+
     async def insert_archival(self, user_id: str, content: str, source: str = "manual") -> None:
         """Insert a long-term archival memory entry."""
         fernet = await self._get_fernet(user_id)
@@ -343,13 +567,26 @@ class MemoryMiddleware:
 
     async def _get_user_debug(self, user_id: str) -> dict[str, str | None]:
         """Load lightweight user debug fields for trace logs."""
+        from app.config.settings import settings  # noqa: PLC0415
+        from app.models import Subscription  # noqa: PLC0415
+
         result = await self._db.execute(select(User).where(User.id == user_id))
         user = result.scalar_one_or_none()
         if user is None:
             return {"tier": None}
-        return {
+
-            "tier": user.tier,
+        sub_result = await self._db.execute(
-        }
+            select(Subscription.tier).where(Subscription.user_id == user_id)
+        )
+        sub_tier: str | None = sub_result.scalar_one_or_none()
+        if sub_tier:
+            tier = sub_tier
+        elif settings.ENV == "dev":
+            tier = "power"
+        else:
+            tier = user.tier or "free"
+
+        return {"tier": tier}
 
     async def _load_core(self, user_id: str, fernet: Fernet) -> dict[str, str]:
         result = await self._db.execute(
@@ -364,14 +601,49 @@ class MemoryMiddleware:
         return out
 
     async def _load_associative(
-        self, user_id: str, message: str, fernet: Fernet
+        self, user_id: str, message: str, fernet: Fernet, *, user_tier: str = "free"
     ) -> list[str]:
         """Load top-k associative memories.
 
-        Production: uses pgvector cosine similarity on the message embedding.
+        Pro+: pgvector cosine similarity on the message embedding (real_embeddings feature).
-        Current implementation: keyword-based fallback (no external embedding call)
+        Free / embedding failure: keyword-ordered fallback (most recent rows).
-        so tests pass without a live OpenAI key.
         """
+        from app.billing.tier_manager import tier_manager  # noqa: PLC0415
+        from app.core.embeddings import embed_text  # noqa: PLC0415
+
+        if tier_manager.check_feature(user_tier, "real_embeddings"):
+            vec = await embed_text(message)
+            if vec is not None:
+                try:
+                    result = await self._db.execute(
+                        select(MemoryAssociative)
+                        .where(
+                            MemoryAssociative.user_id == user_id,
+                            MemoryAssociative.embedding.isnot(None),
+                        )
+                        .order_by(MemoryAssociative.embedding.cosine_distance(vec))
+                        .limit(_ASSOCIATIVE_TOP_K)
+                    )
+                    rows = result.scalars().all()
+                    out: list[str] = []
+                    for row in rows:
+                        plaintext = _safe_decrypt(fernet, row.content_encrypted)
+                        if plaintext is not None:
+                            out.append(plaintext)
+                    logger.info(
+                        "memory: _load_associative user=%s mode=vector hits=%d",
+                        user_id,
+                        len(out),
+                    )
+                    return out
+                except Exception as exc:
+                    logger.warning(
+                        "memory: vector search failed user=%s, falling back to keyword: %s",
+                        user_id,
+                        exc,
+                    )
+
+        # Keyword fallback: most recent rows
         result = await self._db.execute(
             select(MemoryAssociative)
             .where(MemoryAssociative.user_id == user_id)
@@ -379,7 +651,7 @@ class MemoryMiddleware:
             .limit(_ASSOCIATIVE_TOP_K)
         )
         rows = result.scalars().all()
-        out: list[str] = []
+        out = []
         for row in rows:
             plaintext = _safe_decrypt(fernet, row.content_encrypted)
             if plaintext is not None:
@@ -408,6 +680,26 @@ class MemoryMiddleware:
                 out.append(plaintext)
         return out
 
+    async def _load_relational(self, user_id: str, *, user_tier: str = "free") -> list[str]:
+        """Return top-10 relation strings for Pro+ users; empty list for Free."""
+        from app.billing.tier_manager import tier_manager  # noqa: PLC0415
+
+        if not tier_manager.check_feature(user_tier, "relational_memory"):
+            return []
+
+        result = await self._db.execute(
+            select(MemoryRelation)
+            .where(MemoryRelation.user_id == user_id)
+            .order_by(MemoryRelation.confidence.desc())
+            .limit(10)
+        )
+        rows = result.scalars().all()
+        out = [
+            f"{r.subject_label} --{r.predicate}--> {r.object_label}"
+            for r in rows
+        ]
+        return out
+
     async def _load_proactive(self, user_id: str, fernet: Fernet) -> list[str]:
         result = await self._db.execute(
             select(MemoryProactive)

app/core/preprocessors/__init__.py

@@ -0,0 +1,104 @@
+"""Preprocessor registry: detect content type and dispatch to handlers.
+
+Public API
+----------
+detect_content_type(filename, raw_content) -> str
+    Heuristic detection based on file extension and content patterns.
+
+preprocess(content_type, raw_content) -> PreprocessResult
+    Dispatch to the appropriate handler.
+"""
+
+from __future__ import annotations
+
+import re
+
+from app.core.preprocessors.base import PreprocessResult
+
+# ── Heuristics ────────────────────────────────────────────────────────
+
+# Patterns that strongly suggest an email HTML file
+_EMAIL_SIGNALS = re.compile(
+    r"(Subject:|From:|To:|Date:|Sent:|MIME-Version:|Content-Type:\s*text/html)",
+    re.IGNORECASE,
+)
+
+# Patterns that suggest a generic HTML page (not an email)
+_GENERIC_HTML_SIGNALS = re.compile(
+    r"<(nav|main|header|footer|article|section)\b",
+    re.IGNORECASE,
+)
+
+
+def detect_content_type(filename: str, raw_content: str) -> str:
+    """Return a content-type string for the given file.
+
+    Supported types: ``"email_html"``, ``"generic_html"``,
+    ``"plain_text"``, ``"unknown"``.
+    """
+    ext = filename.rsplit(".", 1)[-1].lower() if "." in filename else ""
+
+    if ext == "txt":
+        return "plain_text"
+
+    if ext in ("html", "htm", "eml", "mhtml", "mht"):
+        # Prefer email detection over generic HTML
+        if _EMAIL_SIGNALS.search(raw_content[:4096]):
+            return "email_html"
+        if _GENERIC_HTML_SIGNALS.search(raw_content[:4096]) or "<html" in raw_content[:200].lower():
+            return "generic_html"
+        # .html without clear signals — check for any email header
+        if re.search(r"^(From|To|Subject|Date):", raw_content[:2048], re.MULTILINE | re.IGNORECASE):
+            return "email_html"
+        return "generic_html"
+
+    # Plain text files with email headers
+    if ext in ("", "txt") or not ext:
+        if _EMAIL_SIGNALS.search(raw_content[:4096]):
+            return "email_html"
+
+    # Detect binary content
+    try:
+        raw_content.encode("utf-8")
+    except (UnicodeEncodeError, AttributeError):
+        return "unknown"
+
+    # Non-text bytes heuristic: high ratio of non-printable chars
+    sample = raw_content[:512]
+    non_printable = sum(1 for c in sample if ord(c) < 32 and c not in "\r\n\t")
+    if len(sample) > 0 and non_printable / len(sample) > 0.1:
+        return "unknown"
+
+    return "unknown"
+
+
+# ── Generic fallback handler ──────────────────────────────────────────
+
+def _preprocess_generic(raw_content: str, content_type: str) -> PreprocessResult:
+    """Strip HTML tags if present, return text as-is."""
+    try:
+        from bs4 import BeautifulSoup
+        text = BeautifulSoup(raw_content, "html.parser").get_text(separator="\n")
+    except ImportError:
+        # No BeautifulSoup — strip tags with a simple regex
+        text = re.sub(r"<[^>]+>", "", raw_content)
+
+    text = re.sub(r"\n{3,}", "\n\n", text).strip()
+    return PreprocessResult(content_type=content_type, clean_text=text, metadata={})
+
+
+# ── Dispatch ──────────────────────────────────────────────────────────
+
+def preprocess(content_type: str, raw_content: str) -> PreprocessResult:
+    """Dispatch *raw_content* to the handler registered for *content_type*.
+
+    Falls back to the generic handler for unknown types.
+    """
+    if content_type == "email_html":
+        from app.core.preprocessors.email_html import preprocess_email_html
+        return preprocess_email_html(raw_content)
+
+    return _preprocess_generic(raw_content, content_type)
+
+
+__all__ = ["detect_content_type", "preprocess", "PreprocessResult"]

app/core/preprocessors/base.py

@@ -0,0 +1,25 @@
+"""Base types for the preprocessor system."""
+
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+
+
+@dataclass
+class PreprocessResult:
+    """Output of a preprocessor handler.
+
+    Attributes
+    ----------
+    content_type:
+        The detected content type (e.g. ``"email_html"``, ``"plain_text"``).
+    clean_text:
+        Human-readable text stripped of markup/binary noise.
+    metadata:
+        Dict of extracted metadata (keys vary by handler).
+        Common keys: ``subject``, ``from``, ``to``, ``date``, ``filename``.
+    """
+
+    content_type: str
+    clean_text: str
+    metadata: dict = field(default_factory=dict)

app/core/preprocessors/email_html.py

@@ -0,0 +1,111 @@
+"""Preprocessor for email HTML files.
+
+Handles:
+- HTML stripping via BeautifulSoup
+- Metadata extraction (Subject, From, To, Date)
+- Thread splitting — isolates the latest reply
+"""
+
+from __future__ import annotations
+
+import re
+from typing import TYPE_CHECKING
+
+from app.core.preprocessors.base import PreprocessResult
+
+if TYPE_CHECKING:
+    pass
+
+# ── Thread split markers ──────────────────────────────────────────────
+
+# Matches patterns like:
+#   "On Mon, Apr 7, 2026 at 10:00 AM, Alice <alice@co.com> wrote:"
+#   "-----Original Message-----"
+#   "> " (plain-text quote prefix)
+_THREAD_PATTERNS = [
+    re.compile(r"^On\s+.+wrote\s*:", re.IGNORECASE | re.MULTILINE),
+    re.compile(r"^-{3,}\s*(original message|forwarded message)\s*-{3,}", re.IGNORECASE | re.MULTILINE),
+    re.compile(r"^>{1,}\s+\S", re.MULTILINE),
+    re.compile(r"^From:\s+.+\nSent:\s+", re.IGNORECASE | re.MULTILINE),
+]
+
+# ── Metadata patterns (applied on raw HTML / plain fallback) ──────────
+
+_META_PATTERNS: dict[str, list[re.Pattern]] = {
+    "subject": [
+        re.compile(r"<title>(.+?)</title>", re.IGNORECASE | re.DOTALL),
+        re.compile(r"Subject:\s*(.+)", re.IGNORECASE),
+    ],
+    "from": [
+        re.compile(r'<meta[^>]+name=["\']?from["\']?[^>]+content=["\']([^"\']+)["\']', re.IGNORECASE),
+        re.compile(r"From:\s*(.+)", re.IGNORECASE),
+    ],
+    "to": [
+        re.compile(r'<meta[^>]+name=["\']?to["\']?[^>]+content=["\']([^"\']+)["\']', re.IGNORECASE),
+        re.compile(r"To:\s*(.+)", re.IGNORECASE),
+    ],
+    "date": [
+        re.compile(r'<meta[^>]+name=["\']?date["\']?[^>]+content=["\']([^"\']+)["\']', re.IGNORECASE),
+        re.compile(r"Date:\s*(.+)", re.IGNORECASE),
+        re.compile(r"Sent:\s*(.+)", re.IGNORECASE),
+    ],
+}
+
+
+def _extract_metadata(raw_html: str, text: str) -> dict:
+    """Extract Subject/From/To/Date from raw HTML or plain text."""
+    metadata: dict[str, str] = {}
+    for field, patterns in _META_PATTERNS.items():
+        for pat in patterns:
+            m = pat.search(raw_html) or pat.search(text)
+            if m:
+                metadata[field] = m.group(1).strip()
+                break
+    return metadata
+
+
+def _split_thread(text: str) -> str:
+    """Return only the latest message in a threaded email."""
+    earliest_pos: int | None = None
+    for pat in _THREAD_PATTERNS:
+        m = pat.search(text)
+        if m and (earliest_pos is None or m.start() < earliest_pos):
+            earliest_pos = m.start()
+
+    if earliest_pos is not None and earliest_pos > 0:
+        return text[:earliest_pos].strip()
+    return text.strip()
+
+
+def preprocess_email_html(raw_content: str) -> PreprocessResult:
+    """Strip HTML, extract metadata, split thread from an email HTML file."""
+    try:
+        from bs4 import BeautifulSoup  # lazy import — optional dep
+    except ImportError as exc:
+        raise ImportError(
+            "beautifulsoup4 is required for email_html preprocessing. "
+            "Install it with: pip install beautifulsoup4"
+        ) from exc
+
+    # Parse with lxml if available, fall back to html.parser
+    try:
+        soup = BeautifulSoup(raw_content, "lxml")
+    except Exception:
+        soup = BeautifulSoup(raw_content, "html.parser")
+
+    # Remove noise tags
+    for tag in soup(["style", "script", "head", "noscript"]):
+        tag.decompose()
+
+    clean_text = soup.get_text(separator="\n")
+    # Collapse excessive blank lines
+    clean_text = re.sub(r"\n{3,}", "\n\n", clean_text).strip()
+
+    metadata = _extract_metadata(raw_content, clean_text)
+    latest_message = _split_thread(clean_text)
+
+    return PreprocessResult(
+        content_type="email_html",
+        clean_text=latest_message,
+        metadata=metadata,
+    )

app/integrations/ms_graph.py

@@ -25,7 +25,7 @@ from __future__ import annotations
 
 import logging
 import re
-from datetime import datetime, timedelta, timezone
+from datetime import datetime, timezone
 from typing import Any
 
 import httpx

app/main.py

@@ -16,13 +16,87 @@ from app.api.middleware.sanitizer import SanitizerMiddleware
 from app.config.settings import settings
 
 
+async def _memory_audit_cron_tick() -> None:
+    """Weekly cron: contradiction scan + label canonicalization for all users (Phase 7)."""
+    import logging  # noqa: PLC0415
+    _log = logging.getLogger(__name__)
+    _log.info("memory audit cron tick: starting")
+    try:
+        from app.db import async_session  # noqa: PLC0415
+        from app.core.memory_maintenance import audit_memory  # noqa: PLC0415
+        from app.models import User  # noqa: PLC0415
+        from sqlalchemy import select  # noqa: PLC0415
+
+        async with async_session() as db:
+            result = await db.execute(select(User.id))
+            user_ids: list[str] = list(result.scalars().all())
+
+        for uid in user_ids:
+            try:
+                async with async_session() as db:
+                    await audit_memory(db, uid)
+            except Exception as exc:
+                _log.warning("memory audit cron tick: audit_memory failed user=%s: %s", uid, exc)
+
+        _log.info("memory audit cron tick: done users=%d", len(user_ids))
+    except Exception as exc:
+        _log.warning("memory audit cron tick: failed: %s", exc)
+
+
+async def _memory_cron_tick() -> None:
+    """Hourly cron: drain Free-tier extraction queue + mine proactive patterns for Power+ users."""
+    import logging  # noqa: PLC0415
+    _log = logging.getLogger(__name__)
+    _log.info("memory cron tick: starting")
+    try:
+        from app.db import async_session  # noqa: PLC0415
+        from app.core.memory_maintenance import drain_extraction_queue, mine_proactive_patterns  # noqa: PLC0415
+        from app.billing.tier_manager import tier_manager  # noqa: PLC0415
+        from app.models import User  # noqa: PLC0415
+        from sqlalchemy import select  # noqa: PLC0415
+
+        async with async_session() as db:
+            await drain_extraction_queue(db)
+
+        # mine proactive patterns for every Power+ user
+        async with async_session() as db:
+            result = await db.execute(select(User.id))
+            user_ids: list[str] = list(result.scalars().all())
+
+        for uid in user_ids:
+            try:
+                async with async_session() as db:
+                    tier = await tier_manager.get_tier(uid, db)
+                    if tier_manager.check_feature(tier, "proactive_mining"):
+                        await mine_proactive_patterns(db, uid)
+            except Exception as exc:
+                _log.warning("memory cron tick: mine_proactive_patterns failed user=%s: %s", uid, exc)
+
+        _log.info("memory cron tick: done users=%d", len(user_ids))
+    except Exception as exc:
+        _log.warning("memory cron tick: failed: %s", exc)
+
+
 @asynccontextmanager
 async def lifespan(app: FastAPI):
     # Startup: ensure agent tool modules are loaded.
     import app.agents  # noqa: F401
 
+    scheduler = None
+    if settings.SCHEDULER_ENABLED:
+        from apscheduler.schedulers.asyncio import AsyncIOScheduler  # noqa: PLC0415
+
+        scheduler = AsyncIOScheduler()
+        scheduler.add_job(_memory_cron_tick, "interval", hours=1, id="memory_cron")
+        scheduler.add_job(_memory_audit_cron_tick, "interval", weeks=1, id="memory_audit_cron")
+        scheduler.start()
+        logging.getLogger(__name__).info("memory cron scheduler started (interval=1h)")
+
     yield
 
+    if scheduler is not None:
+        scheduler.shutdown(wait=False)
+
     # Shutdown: dispose SQLAlchemy connection pool
     from app.db import engine
     await engine.dispose()
@@ -30,7 +104,7 @@ async def lifespan(app: FastAPI):
 
 def create_app() -> FastAPI:
     app = FastAPI(
-        title="Adiuva Cloud API",
+        title="AdiuvAI Cloud API",
         version="0.1.0",
         docs_url="/docs" if settings.ENV == "dev" else None,
         redoc_url=None,
@@ -50,17 +124,14 @@ def create_app() -> FastAPI:
     app.add_middleware(SanitizerMiddleware)
     app.add_middleware(TierRateLimitMiddleware)
 
-    from app.api.routes import agents, auth, backup, billing, chat, device_ws, plugins, storage, vectors
+    from app.api.routes import agents, auth, billing, chat, device_ws, memory
 
     app.include_router(auth.router,       prefix="/api/v1")
     app.include_router(chat.router,       prefix="/api/v1")
-    app.include_router(storage.router,    prefix="/api/v1")
-    app.include_router(vectors.router,    prefix="/api/v1")
-    app.include_router(backup.router,     prefix="/api/v1")
-    app.include_router(plugins.router,    prefix="/api/v1")
     app.include_router(billing.router,    prefix="/api/v1")
     app.include_router(agents.router,     prefix="/api/v1")
     app.include_router(device_ws.router,  prefix="/api/v1")
+    app.include_router(memory.router,     prefix="/api/v1")
 
     @app.get("/api/v1/health", tags=["health"])
     async def health() -> dict:

app/marketplace/__init__.py

@@ -1,7 +0,0 @@
-"""Plugin marketplace package.
-
-Three service classes introduced in Step 10:
-  - ``PluginRegistry``  — catalog, submit/approve/reject, install counts
-  - ``ReviewQueue``     — approval workflow + security checklist
-  - ``RevenueShare``    — 70/30 split tracking and Stripe Connect payouts
-"""

app/marketplace/plugin_registry.py

@@ -1,212 +0,0 @@
-"""Plugin catalog registry backed by PostgreSQL.
-
-Maintains the authoritative list of plugins, their review status, and
-aggregate install counts.  All data is persisted in the ``plugins`` table.
-
-Module-level singleton::
-
-    from app.marketplace.plugin_registry import registry
-"""
-
-from __future__ import annotations
-
-import json
-from typing import Any, Literal
-
-from sqlalchemy import select, func
-from sqlalchemy.ext.asyncio import AsyncSession
-
-from app.models import Plugin
-from app.schemas import PluginListResponse, PluginManifest
-
-_PAGE_SIZE = 20
-
-
-def _plugin_to_manifest(p: Plugin) -> PluginManifest:
-    """Convert an ORM ``Plugin`` row to a Pydantic ``PluginManifest``."""
-    try:
-        permissions = json.loads(p.permissions) if p.permissions else []
-    except (json.JSONDecodeError, TypeError):
-        permissions = []
-    return PluginManifest(
-        id=p.id,
-        name=p.name,
-        description=p.description,
-        version=p.version,
-        author=p.author_name,
-        permissions=permissions,
-        category=p.category,
-        price_cents=p.price_cents,
-    )
-
-
-class PluginRegistry:
-    """PostgreSQL-backed plugin catalog.
-
-    All methods accept an ``AsyncSession`` parameter so the calling route
-    controls the session lifecycle.
-    """
-
-    # ── Queries ──────────────────────────────────────────────────────
-
-    async def list_plugins(
-        self,
-        db: AsyncSession,
-        category: str | None = None,
-        query: str | None = None,
-        page: int = 1,
-        sort: Literal["rating", "installs", "newest"] = "newest",
-    ) -> PluginListResponse:
-        """Return a page of approved plugins, optionally filtered and sorted."""
-        base = select(Plugin).where(Plugin.status == "approved")
-
-        if category:
-            base = base.where(Plugin.category == category)
-        if query:
-            pattern = f"%{query}%"
-            base = base.where(
-                Plugin.name.ilike(pattern) | Plugin.description.ilike(pattern)
-            )
-
-        # Count
-        count_q = select(func.count()).select_from(base.subquery())
-        total = (await db.execute(count_q)).scalar_one()
-
-        # Sort
-        if sort == "installs":
-            base = base.order_by(Plugin.install_count.desc())
-        elif sort == "rating":
-            base = base.order_by(Plugin.avg_rating.desc())
-        else:  # newest
-            base = base.order_by(Plugin.created_at.desc())
-
-        base = base.offset((page - 1) * _PAGE_SIZE).limit(_PAGE_SIZE)
-        rows = (await db.execute(base)).scalars().all()
-
-        return PluginListResponse(
-            plugins=[_plugin_to_manifest(r) for r in rows],
-            total=total,
-            page=page,
-        )
-
-    async def get_plugin(self, db: AsyncSession, plugin_id: str) -> dict[str, Any] | None:
-        """Return ``{manifest, status, install_count, avg_rating}`` or ``None``."""
-        result = await db.execute(select(Plugin).where(Plugin.id == plugin_id))
-        p = result.scalar_one_or_none()
-        if p is None:
-            return None
-        return {
-            "manifest": _plugin_to_manifest(p),
-            "status": p.status,
-            "install_count": p.install_count,
-            "avg_rating": p.avg_rating,
-        }
-
-    # ── Mutations ────────────────────────────────────────────────────
-
-    async def submit_plugin(
-        self,
-        db: AsyncSession,
-        manifest: PluginManifest,
-        package_s3_key: str,
-    ) -> str:
-        """Add *manifest* to the catalog with ``status='pending_review'``.
-
-        Returns the plugin_id.  If a plugin with the same id already exists
-        it is overwritten (re-submission after rejection).
-        """
-        plugin_id = manifest.id
-        existing = await db.execute(select(Plugin).where(Plugin.id == plugin_id))
-        row = existing.scalar_one_or_none()
-
-        if row is not None:
-            row.name = manifest.name
-            row.description = manifest.description
-            row.version = manifest.version
-            row.author_name = manifest.author
-            row.category = manifest.category
-            row.price_cents = manifest.price_cents
-            row.permissions = json.dumps(manifest.permissions)
-            row.status = "pending_review"
-            row.s3_package_key = package_s3_key
-            row.rejection_reason = None
-        else:
-            row = Plugin(
-                id=plugin_id,
-                name=manifest.name,
-                description=manifest.description,
-                version=manifest.version,
-                author_name=manifest.author,
-                category=manifest.category,
-                price_cents=manifest.price_cents,
-                permissions=json.dumps(manifest.permissions),
-                status="pending_review",
-                s3_package_key=package_s3_key,
-                install_count=0,
-                avg_rating=0.0,
-            )
-            db.add(row)
-        await db.commit()
-        return plugin_id
-
-    async def approve_plugin(self, db: AsyncSession, plugin_id: str) -> None:
-        """Set *plugin_id* status to ``'approved'``.
-
-        Raises ``KeyError`` if the plugin is not found.
-        """
-        result = await db.execute(select(Plugin).where(Plugin.id == plugin_id))
-        row = result.scalar_one_or_none()
-        if row is None:
-            raise KeyError(f"Plugin not found: {plugin_id}")
-        row.status = "approved"
-        row.rejection_reason = None
-        await db.commit()
-
-    async def reject_plugin(self, db: AsyncSession, plugin_id: str, reason: str) -> None:
-        """Set *plugin_id* status to ``'rejected'`` and record the reason.
-
-        Raises ``KeyError`` if the plugin is not found.
-        """
-        result = await db.execute(select(Plugin).where(Plugin.id == plugin_id))
-        row = result.scalar_one_or_none()
-        if row is None:
-            raise KeyError(f"Plugin not found: {plugin_id}")
-        row.status = "rejected"
-        row.rejection_reason = reason
-        await db.commit()
-
-    async def record_install(self, db: AsyncSession, plugin_id: str) -> None:
-        """Increment the install count for *plugin_id* (no-op if not found)."""
-        result = await db.execute(select(Plugin).where(Plugin.id == plugin_id))
-        row = result.scalar_one_or_none()
-        if row is not None:
-            row.install_count = row.install_count + 1
-            await db.commit()
-
-    async def record_uninstall(self, db: AsyncSession, plugin_id: str) -> None:
-        """Decrement the install count for *plugin_id*, floored at 0."""
-        result = await db.execute(select(Plugin).where(Plugin.id == plugin_id))
-        row = result.scalar_one_or_none()
-        if row is not None:
-            row.install_count = max(0, row.install_count - 1)
-            await db.commit()
-
-    # ── Internal helpers used by ReviewQueue ─────────────────────────
-
-    async def get_pending_entries(self, db: AsyncSession) -> list[dict[str, Any]]:
-        """Return all entries with status='pending_review'."""
-        result = await db.execute(
-            select(Plugin).where(Plugin.status == "pending_review")
-        )
-        rows = result.scalars().all()
-        return [
-            {
-                "manifest": _plugin_to_manifest(r),
-                "submitted_at": int(r.submitted_at.timestamp()) if r.submitted_at else 0,
-            }
-            for r in rows
-        ]
-
-
-# Module-level singleton
-registry = PluginRegistry()

app/marketplace/plugin_review.py

@@ -1,125 +0,0 @@
-"""Plugin review workflow backed by PostgreSQL.
-
-Manages the approval queue for newly submitted plugins and enforces a
-security checklist before any plugin is made visible in the marketplace.
-
-Module-level singleton::
-
-    from app.marketplace.plugin_review import review_queue
-"""
-
-from __future__ import annotations
-
-import re
-from typing import Any, Literal
-
-from sqlalchemy.ext.asyncio import AsyncSession
-
-from app.marketplace.plugin_registry import registry
-from app.models import PluginReview as PluginReviewModel
-from app.schemas import PluginManifest
-
-# ── Security policy ───────────────────────────────────────────────────
-
-ALLOWED_PERMISSIONS: frozenset[str] = frozenset(
-    {
-        "read:tasks",
-        "write:tasks",
-        "read:projects",
-        "write:projects",
-        "read:notes",
-        "write:notes",
-        "read:timelines",
-        "write:timelines",
-        "read:calendar",
-        "write:calendar",
-    }
-)
-
-_PLUGIN_ID_RE = re.compile(r"^[a-z0-9-]+$")
-
-
-def validate_manifest(manifest: PluginManifest) -> None:
-    """Enforce the plugin security checklist.
-
-    Raises:
-        ``ValueError`` on the first violation found.  Callers should catch
-        this and return HTTP 422 / reject the submission.
-
-    Checks:
-      1. Plugin id matches ``^[a-z0-9-]+$``
-      2. All declared permissions are in ``ALLOWED_PERMISSIONS``
-      3. No manifest field contains raw binary data
-    """
-    if not _PLUGIN_ID_RE.match(manifest.id):
-        raise ValueError(
-            f"Invalid plugin id format: '{manifest.id}'. "
-            "Only lowercase letters, digits, and hyphens are allowed."
-        )
-
-    for perm in manifest.permissions:
-        if perm not in ALLOWED_PERMISSIONS:
-            raise ValueError(
-                f"Unknown permission: '{perm}'. "
-                f"Allowed permissions: {sorted(ALLOWED_PERMISSIONS)}"
-            )
-
-    for field_name, value in manifest.model_dump().items():
-        if isinstance(value, (bytes, bytearray)):
-            raise ValueError(
-                f"Binary content is not allowed in manifest field '{field_name}'."
-            )
-
-
-class ReviewQueue:
-    """Approval queue for pending plugin submissions.
-
-    Delegates status changes to the shared ``PluginRegistry`` singleton.
-    Review records are persisted in the ``plugin_reviews`` table.
-    """
-
-    async def get_pending(self, db: AsyncSession) -> list[dict[str, Any]]:
-        """Return all plugins currently awaiting review.
-
-        Each item is ``{plugin_id, manifest, submitted_at}``.
-        """
-        entries = await registry.get_pending_entries(db)
-        return [
-            {
-                "plugin_id": e["manifest"].id,
-                "manifest": e["manifest"],
-                "submitted_at": e["submitted_at"],
-            }
-            for e in entries
-        ]
-
-    async def submit_review(
-        self,
-        db: AsyncSession,
-        plugin_id: str,
-        reviewer_id: str,
-        decision: Literal["approved", "rejected"],
-        notes: str = "",
-    ) -> None:
-        """Record a review decision and update the plugin's status.
-
-        Raises:
-            ``KeyError`` if *plugin_id* is not found in the registry.
-        """
-        if decision == "approved":
-            await registry.approve_plugin(db, plugin_id)
-        else:
-            await registry.reject_plugin(db, plugin_id, reason=notes)
-
-        review = PluginReviewModel(
-            plugin_id=plugin_id,
-            reviewer_id=reviewer_id,
-            decision=decision,
-            notes=notes,
-        )
-        db.add(review)
-        await db.commit()
-
-
-# Module-level singleton
-review_queue = ReviewQueue()

app/marketplace/revenue_share.py

@@ -1,233 +0,0 @@
-"""Revenue share tracking and Stripe Connect payouts backed by PostgreSQL.
-
-Records every plugin installation as a revenue event and facilitates
-70 % / 30 % payouts to developers via Stripe Connect.  Data is persisted
-in the ``revenue_events`` table.
-
-Module-level singleton::
-
-    from app.marketplace.revenue_share import revenue_share
-"""
-
-from __future__ import annotations
-
-import logging
-from datetime import datetime, timezone
-from typing import Any
-
-import stripe as stripe_lib
-from sqlalchemy import extract, func, select
-from sqlalchemy.ext.asyncio import AsyncSession
-
-from app.config.settings import settings
-from app.marketplace.plugin_registry import registry
-from app.models import Plugin, RevenueEvent
-
-logger = logging.getLogger(__name__)
-
-# ── Revenue split constants ───────────────────────────────────────────
-
-DEVELOPER_SHARE: float = 0.70
-PLATFORM_SHARE: float = 0.30
-
-
-class RevenueShare:
-    """Records installation revenue events and coordinates developer payouts.
-
-    Stripe Connect calls are gracefully stubbed when ``STRIPE_SECRET_KEY``
-    is not configured, consistent with the rest of the billing layer.
-    """
-
-    # ── Helpers ──────────────────────────────────────────────────────
-
-    @staticmethod
-    def _stripe_configured() -> bool:
-        return bool(settings.STRIPE_SECRET_KEY)
-
-    @staticmethod
-    def _stripe() -> Any:
-        stripe_lib.api_key = settings.STRIPE_SECRET_KEY
-        return stripe_lib
-
-    # ── Core operations ──────────────────────────────────────────────
-
-    async def record_install(
-        self,
-        db: AsyncSession,
-        plugin_id: str,
-        user_id: str,
-        amount_cents: int,
-    ) -> None:
-        """Record a plugin installation and trigger a Stripe Connect charge if paid.
-
-        For free plugins (``amount_cents == 0``) no payment is initiated but
-        the event is still recorded for analytics.
-
-        For paid plugins the developer receives 70 % via a Stripe Connect
-        destination charge.  If Stripe is not configured or the charge fails
-        the installation still succeeds (the event is recorded and the install
-        count is incremented) — a warning is logged for monitoring.
-        """
-        developer_share_cents = int(amount_cents * DEVELOPER_SHARE)
-        stripe_transfer_id: str | None = None
-
-        if amount_cents > 0 and self._stripe_configured():
-            # Look up the plugin's author Stripe account from the DB
-            result = await db.execute(select(Plugin).where(Plugin.id == plugin_id))
-            plugin_row = result.scalar_one_or_none()
-            developer_stripe_account: str | None = None
-            if plugin_row and plugin_row.author_id:
-                # Future: look up user.stripe_connect_account_id
-                developer_stripe_account = None  # no real account yet
-
-            if developer_stripe_account:
-                try:
-                    s = self._stripe()
-                    transfer = s.Transfer.create(
-                        amount=developer_share_cents,
-                        currency="eur",
-                        destination=developer_stripe_account,
-                        description=f"Revenue share for plugin {plugin_id}",
-                        metadata={"plugin_id": plugin_id, "user_id": user_id},
-                    )
-                    stripe_transfer_id = transfer["id"]
-                except Exception as exc:
-                    logger.warning(
-                        "Stripe Connect transfer failed for plugin %s: %s",
-                        plugin_id,
-                        exc,
-                    )
-            else:
-                logger.debug(
-                    "No Stripe account on file for plugin %s developer; "
-                    "skipping transfer.",
-                    plugin_id,
-                )
-
-        event = RevenueEvent(
-            plugin_id=plugin_id,
-            user_id=user_id,
-            amount_cents=amount_cents,
-            developer_share_cents=developer_share_cents,
-            stripe_transfer_id=stripe_transfer_id,
-        )
-        db.add(event)
-        await db.commit()
-
-        await registry.record_install(db, plugin_id)
-
-    async def get_earnings(
-        self,
-        db: AsyncSession,
-        developer_id: str,
-        period: str | None = None,
-    ) -> dict[str, Any]:
-        """Return aggregated earnings for *developer_id*.
-
-        ``period`` is an optional ``YYYY-MM`` string to restrict the window.
-
-        Returns::
-
-            {
-                "developer_id": str,
-                "period": str | None,
-                "total_installs": int,
-                "total_revenue_cents": int,
-                "developer_share_cents": int,
-            }
-        """
-        # Find plugin ids belonging to this developer (by author_name match)
-        plugin_q = select(Plugin.id).where(Plugin.author_name == developer_id)
-        plugin_result = await db.execute(plugin_q)
-        developer_plugin_ids = [row[0] for row in plugin_result.all()]
-
-        if not developer_plugin_ids:
-            return {
-                "developer_id": developer_id,
-                "period": period,
-                "total_installs": 0,
-                "total_revenue_cents": 0,
-                "developer_share_cents": 0,
-            }
-
-        query = select(
-            func.count().label("total_installs"),
-            func.coalesce(func.sum(RevenueEvent.amount_cents), 0).label("total_revenue"),
-            func.coalesce(func.sum(RevenueEvent.developer_share_cents), 0).label("dev_share"),
-        ).where(RevenueEvent.plugin_id.in_(developer_plugin_ids))
-
-        if period:
-            # Filter by YYYY-MM: extract year and month from created_at
-            try:
-                year, month = period.split("-")
-                query = query.where(
-                    extract("year", RevenueEvent.created_at) == int(year),
-                    extract("month", RevenueEvent.created_at) == int(month),
-                )
-            except ValueError:
-                pass  # invalid period format — return all
-
-        result = await db.execute(query)
-        row = result.one()
-
-        return {
-            "developer_id": developer_id,
-            "period": period,
-            "total_installs": row.total_installs,
-            "total_revenue_cents": row.total_revenue,
-            "developer_share_cents": row.dev_share,
-        }
-
-    async def payout_developer(self, db: AsyncSession, plugin_id: str, period: str) -> None:
-        """Aggregate unpaid revenue for *period* and issue a Stripe Transfer.
-
-        Marks processed events with ``paid_at`` timestamp.
-        Stubs gracefully when Stripe is not configured.
-        """
-        try:
-            year, month = period.split("-")
-            year_int, month_int = int(year), int(month)
-        except ValueError:
-            logger.warning("Invalid period format: %s", period)
-            return
-
-        result = await db.execute(
-            select(RevenueEvent).where(
-                RevenueEvent.plugin_id == plugin_id,
-                RevenueEvent.paid_at.is_(None),
-                extract("year", RevenueEvent.created_at) == year_int,
-                extract("month", RevenueEvent.created_at) == month_int,
-            )
-        )
-        unpaid = list(result.scalars().all())
-
-        total_dev_share = sum(e.developer_share_cents for e in unpaid)
-        if total_dev_share <= 0 or not unpaid:
-            logger.debug("Nothing to pay out for plugin %s in period %s", plugin_id, period)
-            return
-
-        if self._stripe_configured():
-            plugin_result = await db.execute(select(Plugin).where(Plugin.id == plugin_id))
-            plugin_row = plugin_result.scalar_one_or_none()
-            developer_stripe_account: str | None = None  # Future: fetch from DB
-            if plugin_row and developer_stripe_account:
-                try:
-                    s = self._stripe()
-                    s.Transfer.create(
-                        amount=total_dev_share,
-                        currency="eur",
-                        destination=developer_stripe_account,
-                        description=f"Payout for plugin {plugin_id} period {period}",
-                    )
-                except Exception as exc:
-                    logger.warning("Payout transfer failed for plugin %s: %s", plugin_id, exc)
-                    return
-
-        paid_ts = datetime.now(timezone.utc)
-        for event in unpaid:
-            event.paid_at = paid_ts
-        await db.commit()
-
-
-# Module-level singleton
-revenue_share = RevenueShare()

app/models.py

@@ -1,23 +1,20 @@
 """SQLAlchemy ORM models for all persistent tables.
 
-Only auth, billing, storage metadata, and marketplace data live here.
+Only auth, billing, agent config, and memory data live here.
-User content (notes, tasks, etc.) is NEVER persisted server-side —
+User content (notes, tasks, etc.) lives exclusively on the client.
-it lives in E2E-encrypted blobs in S3, referenced by storage_records.
 
 Table inventory:
   users               — account credentials + tier
   refresh_tokens      — hashed refresh token store
   subscriptions       — Stripe subscription records
-  storage_records     — S3 blob metadata (no plaintext)
+  local_agent_configs — per-device batch agent configs
-  backup_metadata     — encrypted backup manifests
+  cloud_agent_configs — OAuth-backed cloud agent configs
-  plugins             — marketplace plugin catalog
+  agent_run_logs      — execution history for all agents
-  plugin_installations — per-user install records
-  plugin_reviews      — admin review decisions
-  revenue_events      — Stripe Connect 70/30 split ledger
   memory_core         — per-user persistent key/value preferences (encrypted)
   memory_associative  — per-user semantic memory with embeddings (encrypted)
   memory_episodic     — per-user session summaries (encrypted)
   memory_proactive    — per-user behavioral patterns (encrypted)
+  memory_relations    — per-user entity/relation graph (Mem0g-light, Phase 3)
 """
 
 from __future__ import annotations
@@ -25,8 +22,8 @@ from __future__ import annotations
 import uuid
 from datetime import datetime, timezone
 
+from pgvector.sqlalchemy import Vector
 from sqlalchemy import (
-    BigInteger,
     Boolean,
     DateTime,
     Enum,
@@ -34,9 +31,9 @@ from sqlalchemy import (
     ForeignKey,
     Integer,
     JSON,
+    LargeBinary,
     String,
     Text,
-    UniqueConstraint,
     Uuid,
     func,
 )
@@ -58,8 +55,6 @@ def _now() -> datetime:
 # ── Enum types ────────────────────────────────────────────────────────────
 
 TierEnum = Enum("free", "pro", "power", "team", name="billing_tier")
-PluginStatusEnum = Enum("pending_review", "approved", "rejected", name="plugin_status")
-ReviewDecisionEnum = Enum("approved", "rejected", name="review_decision")
 AgentTypeEnum = Enum("local", "cloud", name="agent_type")
 AgentStatusEnum = Enum("running", "success", "error", "partial", name="agent_run_status")
 CloudProviderEnum = Enum("gmail", "teams", "outlook", name="cloud_provider")
@@ -77,7 +72,8 @@ class User(Base):
     email: Mapped[str] = mapped_column(String(255), unique=True, nullable=False, index=True)
     name: Mapped[str | None] = mapped_column(String(100), nullable=True)
     surname: Mapped[str | None] = mapped_column(String(100), nullable=True)
-    password_hash: Mapped[str] = mapped_column(String(255), nullable=False)
+    password_hash: Mapped[str | None] = mapped_column(String(255), nullable=True)
+    avatar_url: Mapped[str | None] = mapped_column(Text, nullable=True)
     tier: Mapped[str] = mapped_column(TierEnum, nullable=False, default="free")
     stripe_customer_id: Mapped[str | None] = mapped_column(String(255), nullable=True)
     # Per-user Fernet key (base64-urlsafe, 44 chars). Generated on registration.
@@ -86,6 +82,9 @@ class User(Base):
     created_at: Mapped[datetime] = mapped_column(
         DateTime(timezone=True), nullable=False, server_default=func.now()
     )
+    onboarding_completed_at: Mapped[datetime | None] = mapped_column(
+        DateTime(timezone=True), nullable=True, default=None
+    )
     updated_at: Mapped[datetime] = mapped_column(
         DateTime(timezone=True), nullable=False, server_default=func.now(), onupdate=func.now()
     )
@@ -96,6 +95,9 @@ class User(Base):
     subscription: Mapped[Subscription | None] = relationship(
         back_populates="user", uselist=False, cascade="all, delete-orphan"
     )
+    oauth_accounts: Mapped[list[OAuthAccount]] = relationship(
+        back_populates="user", cascade="all, delete-orphan"
+    )
 
 
 class RefreshToken(Base):
@@ -116,6 +118,25 @@ class RefreshToken(Base):
     user: Mapped[User] = relationship(back_populates="refresh_tokens")
 
 
+class OAuthAccount(Base):
+    __tablename__ = "oauth_accounts"
+
+    id: Mapped[str] = mapped_column(
+        Uuid(as_uuid=False), primary_key=True, default=_uuid
+    )
+    user_id: Mapped[str] = mapped_column(
+        Uuid(as_uuid=False), ForeignKey("users.id", ondelete="CASCADE"), nullable=False, index=True
+    )
+    provider: Mapped[str] = mapped_column(String(50), nullable=False)
+    provider_user_id: Mapped[str] = mapped_column(String(255), nullable=False)
+    provider_email: Mapped[str | None] = mapped_column(String(255), nullable=True)
+    created_at: Mapped[datetime] = mapped_column(
+        DateTime(timezone=True), nullable=False, server_default=func.now()
+    )
+
+    user: Mapped[User] = relationship(back_populates="oauth_accounts")
+
+
 class Subscription(Base):
     __tablename__ = "subscriptions"
 
@@ -137,151 +158,6 @@ class Subscription(Base):
     user: Mapped[User] = relationship(back_populates="subscription")
 
 
-class StorageRecord(Base):
-    __tablename__ = "storage_records"
-
-    id: Mapped[str] = mapped_column(
-        Uuid(as_uuid=False), primary_key=True, default=_uuid
-    )
-    user_id: Mapped[str] = mapped_column(
-        Uuid(as_uuid=False), ForeignKey("users.id", ondelete="CASCADE"), nullable=False, index=True
-    )
-    table_name: Mapped[str] = mapped_column(String(100), nullable=False)
-    s3_key: Mapped[str] = mapped_column(String(500), nullable=False)
-    checksum: Mapped[str] = mapped_column(String(64), nullable=False)
-    size_bytes: Mapped[int] = mapped_column(Integer, nullable=False)
-    created_at: Mapped[datetime] = mapped_column(
-        DateTime(timezone=True), nullable=False, server_default=func.now()
-    )
-    updated_at: Mapped[datetime] = mapped_column(
-        DateTime(timezone=True), nullable=False, server_default=func.now(), onupdate=func.now()
-    )
-
-
-class BackupMetadata(Base):
-    __tablename__ = "backup_metadata"
-
-    id: Mapped[str] = mapped_column(
-        Uuid(as_uuid=False), primary_key=True, default=_uuid
-    )
-    user_id: Mapped[str] = mapped_column(
-        Uuid(as_uuid=False), ForeignKey("users.id", ondelete="CASCADE"), nullable=False, index=True
-    )
-    s3_key: Mapped[str] = mapped_column(String(500), nullable=False)
-    version: Mapped[int] = mapped_column(Integer, nullable=False)
-    timestamp: Mapped[int] = mapped_column(BigInteger, nullable=False)
-    checksum: Mapped[str] = mapped_column(String(64), nullable=False)
-    size_bytes: Mapped[int] = mapped_column(Integer, nullable=False)
-    created_at: Mapped[datetime] = mapped_column(
-        DateTime(timezone=True), nullable=False, server_default=func.now()
-    )
-
-
-class Plugin(Base):
-    __tablename__ = "plugins"
-
-    id: Mapped[str] = mapped_column(String(255), primary_key=True)
-    name: Mapped[str] = mapped_column(String(255), nullable=False)
-    description: Mapped[str] = mapped_column(Text, nullable=False, default="")
-    version: Mapped[str] = mapped_column(String(50), nullable=False, default="1.0.0")
-    # nullable until developer account system is built
-    author_id: Mapped[str | None] = mapped_column(
-        Uuid(as_uuid=False), ForeignKey("users.id", ondelete="SET NULL"), nullable=True
-    )
-    author_name: Mapped[str] = mapped_column(String(255), nullable=False, default="")
-    category: Mapped[str] = mapped_column(String(100), nullable=False, default="")
-    price_cents: Mapped[int] = mapped_column(Integer, nullable=False, default=0)
-    permissions: Mapped[str] = mapped_column(Text, nullable=False, default="[]")  # JSON list
-    status: Mapped[str] = mapped_column(PluginStatusEnum, nullable=False, default="pending_review")
-    s3_package_key: Mapped[str | None] = mapped_column(String(500), nullable=True)
-    install_count: Mapped[int] = mapped_column(Integer, nullable=False, default=0)
-    avg_rating: Mapped[float] = mapped_column(Float, nullable=False, default=0.0)
-    rejection_reason: Mapped[str | None] = mapped_column(Text, nullable=True)
-    submitted_at: Mapped[datetime] = mapped_column(
-        DateTime(timezone=True), nullable=False, server_default=func.now()
-    )
-    created_at: Mapped[datetime] = mapped_column(
-        DateTime(timezone=True), nullable=False, server_default=func.now()
-    )
-
-    installations: Mapped[list[PluginInstallation]] = relationship(
-        back_populates="plugin", cascade="all, delete-orphan"
-    )
-    reviews: Mapped[list[PluginReview]] = relationship(
-        back_populates="plugin", cascade="all, delete-orphan"
-    )
-    revenue_events: Mapped[list[RevenueEvent]] = relationship(
-        back_populates="plugin", cascade="all, delete-orphan"
-    )
-
-
-class PluginInstallation(Base):
-    __tablename__ = "plugin_installations"
-    __table_args__ = (UniqueConstraint("plugin_id", "user_id", name="uq_plugin_user"),)
-
-    id: Mapped[str] = mapped_column(
-        Uuid(as_uuid=False), primary_key=True, default=_uuid
-    )
-    plugin_id: Mapped[str] = mapped_column(
-        String(255), ForeignKey("plugins.id", ondelete="CASCADE"), nullable=False, index=True
-    )
-    user_id: Mapped[str] = mapped_column(
-        Uuid(as_uuid=False), ForeignKey("users.id", ondelete="CASCADE"), nullable=False, index=True
-    )
-    installed_at: Mapped[datetime] = mapped_column(
-        DateTime(timezone=True), nullable=False, server_default=func.now()
-    )
-
-    plugin: Mapped[Plugin] = relationship(back_populates="installations")
-
-
-class PluginReview(Base):
-    __tablename__ = "plugin_reviews"
-
-    id: Mapped[str] = mapped_column(
-        Uuid(as_uuid=False), primary_key=True, default=_uuid
-    )
-    plugin_id: Mapped[str] = mapped_column(
-        String(255), ForeignKey("plugins.id", ondelete="CASCADE"), nullable=False, index=True
-    )
-    reviewer_id: Mapped[str | None] = mapped_column(
-        Uuid(as_uuid=False), ForeignKey("users.id", ondelete="SET NULL"), nullable=True
-    )
-    decision: Mapped[str] = mapped_column(ReviewDecisionEnum, nullable=False)
-    notes: Mapped[str | None] = mapped_column(Text, nullable=True)
-    reviewed_at: Mapped[datetime] = mapped_column(
-        DateTime(timezone=True), nullable=False, server_default=func.now()
-    )
-    created_at: Mapped[datetime] = mapped_column(
-        DateTime(timezone=True), nullable=False, server_default=func.now()
-    )
-
-    plugin: Mapped[Plugin] = relationship(back_populates="reviews")
-
-
-class RevenueEvent(Base):
-    __tablename__ = "revenue_events"
-
-    id: Mapped[str] = mapped_column(
-        Uuid(as_uuid=False), primary_key=True, default=_uuid
-    )
-    plugin_id: Mapped[str] = mapped_column(
-        String(255), ForeignKey("plugins.id", ondelete="CASCADE"), nullable=False, index=True
-    )
-    user_id: Mapped[str] = mapped_column(
-        Uuid(as_uuid=False), ForeignKey("users.id", ondelete="CASCADE"), nullable=False, index=True
-    )
-    amount_cents: Mapped[int] = mapped_column(Integer, nullable=False, default=0)
-    developer_share_cents: Mapped[int] = mapped_column(Integer, nullable=False, default=0)
-    stripe_transfer_id: Mapped[str | None] = mapped_column(String(255), nullable=True)
-    paid_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True), nullable=True)
-    created_at: Mapped[datetime] = mapped_column(
-        DateTime(timezone=True), nullable=False, server_default=func.now()
-    )
-
-    plugin: Mapped[Plugin] = relationship(back_populates="revenue_events")
-
-
 class LocalAgentConfig(Base):
     __tablename__ = "local_agent_configs"
 
@@ -296,6 +172,7 @@ class LocalAgentConfig(Base):
     directory_paths: Mapped[list] = mapped_column(JSON, nullable=False, default=list)
     data_types: Mapped[list] = mapped_column(JSON, nullable=False, default=list)
     prompt_template: Mapped[str] = mapped_column(Text, nullable=False, default="")
+    agent_config: Mapped[dict | None] = mapped_column(JSON, nullable=True)
     file_extensions: Mapped[list] = mapped_column(JSON, nullable=False, default=list)
     schedule_cron: Mapped[str] = mapped_column(String(100), nullable=False, default="0 */6 * * *")
     enabled: Mapped[bool] = mapped_column(Boolean, nullable=False, default=True)
@@ -425,8 +302,8 @@ class MemoryAssociative(Base):
         nullable=False, index=True,
     )
     content_encrypted: Mapped[str] = mapped_column(Text, nullable=False)
-    # JSON-encoded float list in SQLite tests; vector(1536) in Postgres via migration.
+    # vector(1536) via pgvector; SQLite tests use NULL embeddings so no dialect issue.
-    embedding: Mapped[list | None] = mapped_column(JSON, nullable=True)
+    embedding: Mapped[list | None] = mapped_column(Vector(1536), nullable=True)
     entity_type: Mapped[str | None] = mapped_column(String(100), nullable=True)
     entity_id: Mapped[str | None] = mapped_column(String(255), nullable=True)
     updated_at: Mapped[datetime] = mapped_column(
@@ -474,3 +351,85 @@ class MemoryProactive(Base):
     created_at: Mapped[datetime] = mapped_column(
         DateTime(timezone=True), nullable=False, server_default=func.now()
     )
+
+
+class ExtractionQueue(Base):
+    """Batch extraction queue for Free-tier users (Phase 2).
+
+    Pro/Power/Team users get realtime asyncio.create_task() extraction.
+    Free users get a queue row here; a daily cron (Phase 5) drains it.
+    """
+
+    __tablename__ = "extraction_queue"
+
+    id: Mapped[str] = mapped_column(Uuid(as_uuid=False), primary_key=True, default=_uuid)
+    user_id: Mapped[str] = mapped_column(
+        Uuid(as_uuid=False), ForeignKey("users.id", ondelete="CASCADE"),
+        nullable=False, index=True,
+    )
+    episode_id: Mapped[str | None] = mapped_column(
+        Uuid(as_uuid=False), nullable=True,
+    )
+    created_at: Mapped[datetime] = mapped_column(
+        DateTime(timezone=True), nullable=False, server_default=func.now()
+    )
+
+
+class MemoryRelation(Base):
+    """Per-user entity/relation graph row (Mem0g-light, Phase 3).
+
+    subject_label/object_label are plaintext entity identifiers (not user content).
+    notes_encrypted is optional Fernet-encrypted per-user commentary.
+    confidence in [0.0, 1.0] — decays 5 % per 30 days since last_confirmed_at.
+    """
+
+    __tablename__ = "memory_relations"
+
+    id: Mapped[str] = mapped_column(Uuid(as_uuid=False), primary_key=True, default=_uuid)
+    user_id: Mapped[str] = mapped_column(
+        Uuid(as_uuid=False), ForeignKey("users.id", ondelete="CASCADE"),
+        nullable=False, index=True,
+    )
+    subject_label: Mapped[str] = mapped_column(String(128), nullable=False)
+    subject_type: Mapped[str] = mapped_column(String(32), nullable=False)
+    predicate: Mapped[str] = mapped_column(String(64), nullable=False)
+    object_label: Mapped[str] = mapped_column(String(128), nullable=False)
+    object_type: Mapped[str] = mapped_column(String(32), nullable=False)
+    confidence: Mapped[float] = mapped_column(Float, nullable=False, default=0.7)
+    source_episode_id: Mapped[str | None] = mapped_column(
+        Uuid(as_uuid=False),
+        ForeignKey("memory_episodic.id", ondelete="SET NULL"),
+        nullable=True,
+    )
+    notes_encrypted: Mapped[bytes | None] = mapped_column(LargeBinary, nullable=True)
+    created_at: Mapped[datetime] = mapped_column(
+        DateTime(timezone=True), nullable=False, server_default=func.now()
+    )
+    updated_at: Mapped[datetime] = mapped_column(
+        DateTime(timezone=True), nullable=False, server_default=func.now(), onupdate=func.now()
+    )
+    last_confirmed_at: Mapped[datetime | None] = mapped_column(
+        DateTime(timezone=True), nullable=True
+    )
+
+
+class Plugin(Base):
+    """Plugin marketplace catalog entry."""
+
+    __tablename__ = "plugins"
+
+    id: Mapped[str] = mapped_column(String(255), primary_key=True)
+    name: Mapped[str] = mapped_column(String(255), nullable=False)
+    description: Mapped[str] = mapped_column(Text, nullable=False)
+    version: Mapped[str] = mapped_column(String(50), nullable=False)
+    author_name: Mapped[str] = mapped_column(String(255), nullable=False)
+    category: Mapped[str] = mapped_column(String(100), nullable=False)
+    price_cents: Mapped[int] = mapped_column(Integer, nullable=False, default=0)
+    permissions: Mapped[str] = mapped_column(Text, nullable=False, default="[]")
+    status: Mapped[str] = mapped_column(String(50), nullable=False, default="pending")
+    s3_package_key: Mapped[str | None] = mapped_column(String(500), nullable=True)
+    install_count: Mapped[int] = mapped_column(Integer, nullable=False, default=0)
+    avg_rating: Mapped[float] = mapped_column(Float, nullable=False, default=0.0)
+    created_at: Mapped[datetime] = mapped_column(
+        DateTime(timezone=True), nullable=False, server_default=func.now()
+    )

app/schemas.py

@@ -30,6 +30,16 @@ class UserProfile(BaseModel):
     name: str | None = None
     surname: str | None = None
     tier: BillingTier
+    avatar_url: str | None = None
+    has_password: bool = True
+    onboarding_completed_at: int | None = None  # epoch ms, null = not onboarded
+    memory: dict[str, str] = Field(default_factory=dict)  # decrypted core memory k/v
+
+
+class OAuthAccountInfo(BaseModel):
+    provider: str
+    provider_email: str | None = None
+    created_at: int  # epoch ms
 
 
 # ── Chat ─────────────────────────────────────────────────────────────
@@ -50,88 +60,6 @@ class ChatResponse(BaseModel):
     response: str
 
 
-# ── Backup ───────────────────────────────────────────────────────────
-
-class BackupMetadata(BaseModel):
-    version: int
-    timestamp: int
-    checksum: str
-    chunk_count: int
-
-
-# ── Cloud Storage (E2E encrypted blobs) ──────────────────────────────
-
-class StorageRecord(BaseModel):
-    id: str
-    user_id: str
-    table: str
-    blob: bytes
-    checksum: str
-    created_at: int
-    updated_at: int
-
-
-class StorageRecordCreate(BaseModel):
-    table: str
-    blob: bytes
-    checksum: str
-
-
-class StorageRecordUpdate(BaseModel):
-    blob: bytes
-    checksum: str
-
-
-# ── Cloud Vector Store (E2E encrypted vectors) ────────────────────────
-
-class VectorItem(BaseModel):
-    id: str
-    blob: bytes   # encrypted vector + metadata — backend never decrypts
-    checksum: str
-
-
-class VectorUpsertRequest(BaseModel):
-    vectors: list[VectorItem]
-
-
-class VectorSearchRequest(BaseModel):
-    query_blob: bytes   # encrypted query — backend never decrypts
-    top_k: int = 10
-
-
-class VectorSearchResult(BaseModel):
-    id: str
-    score: float
-    blob: bytes
-
-
-class VectorSearchResponse(BaseModel):
-    results: list[VectorSearchResult]
-
-
-# ── Plugin Marketplace ────────────────────────────────────────────────
-
-class PluginManifest(BaseModel):
-    id: str
-    name: str
-    description: str
-    version: str
-    author: str
-    permissions: list[str]
-    category: str
-    price_cents: int = 0
-
-
-class PluginListResponse(BaseModel):
-    plugins: list[PluginManifest]
-    total: int
-    page: int
-
-
-class PluginInstallRequest(BaseModel):
-    plugin_id: str
-
-
 # ── WebSocket Frame Protocol ──────────────────────────────────────────
 
 class WsFrameType(str, Enum):
@@ -273,6 +201,27 @@ class WsFloatingDomain(BaseModel):
     domain: WsDomain
 
 
+# ── Agent Config V2 ───────────────────────────────────────────────────
+
+
+class ContentTypeConfig(BaseModel):
+    """Per-type extraction config produced by the journey chatbot."""
+
+    id: str
+    label: str = ""
+    detection_hint: str = ""
+    preprocessing: str = "generic"  # handler name: "email_html", "plain_text", ...
+    extraction_prompt: str
+
+
+class AgentConfig(BaseModel):
+    """Structured agent configuration (replaces freeform prompt_template)."""
+
+    content_types: list[ContentTypeConfig] = []
+    global_rules: list[str] = []
+    data_types: list[str] = []
+
+
 # ── Agent Catalog ─────────────────────────────────────────────────────
 
 class AgentCatalogItem(BaseModel):
@@ -297,10 +246,11 @@ class AgentTriggerRequest(BaseModel):
     device_id: str = Field(default="")
     agent_id: str | None = None  # FE stable agent ID (electron-store UUID)
     what_to_extract: list[str] = Field(min_length=1)
-    actions_by_type: dict[str, list[str]] | None = None
     batch_interval: str = Field(min_length=1)
-    custom_agent_prompt: str = Field(min_length=1)
+    custom_agent_prompt: str | None = None
+    agent_config: dict | None = None
     active_agents: int = Field(ge=0, default=0)
+    last_run_at: int | None = None  # epoch ms from FE — enables incremental scanning
 
 
 # ── Agent Run Log ─────────────────────────────────────────────────────

app/storage/__init__.py

@@ -1 +0,0 @@
-"""Cloud storage layer — E2E encrypted blobs and vectors."""

app/storage/blob_store.py

@@ -1,106 +0,0 @@
-"""S3-backed store for E2E-encrypted blobs.
-
-Keys are structured as ``{user_id}/{table}/{record_id}``.
-The backend never inspects blob content — it stores and retrieves opaque bytes.
-"""
-
-from __future__ import annotations
-
-from typing import Any
-
-import boto3
-
-from app.config.settings import settings
-
-
-class BlobStore:
-    """Thin wrapper around boto3 S3.
-
-    All blobs must be E2E encrypted by the client before upload.
-    The backend adds SSE-S3 as an extra layer of at-rest encryption
-    but cannot decrypt the inner client-side payload.
-    """
-
-    def _client(self) -> Any:
-        kwargs: dict[str, Any] = {
-            "region_name": settings.S3_REGION,
-            "aws_access_key_id": settings.AWS_ACCESS_KEY_ID,
-            "aws_secret_access_key": settings.AWS_SECRET_ACCESS_KEY,
-        }
-        if settings.S3_ENDPOINT_URL and isinstance(settings.S3_ENDPOINT_URL, str):
-            kwargs["endpoint_url"] = settings.S3_ENDPOINT_URL
-        return boto3.client("s3", **kwargs)
-
-    @staticmethod
-    def _key(user_id: str, table: str, record_id: str) -> str:
-        return f"{user_id}/{table}/{record_id}"
-
-    async def upload(
-        self,
-        user_id: str,
-        table: str,
-        record_id: str,
-        blob: bytes,
-        checksum: str,
-    ) -> str:
-        """Store *blob* in S3 and return the S3 key.
-
-        Args:
-            user_id:   Owner of the blob (used as key prefix).
-            table:     Logical table name (e.g. ``"tasks"``).
-            record_id: Record UUID.
-            blob:      Raw bytes (pre-encrypted by client).
-            checksum:  SHA-256 hex digest supplied by the client; stored as
-                       object metadata for download-time verification.
-
-        Returns:
-            The S3 key under which the blob was stored.
-        """
-        key = self._key(user_id, table, record_id)
-        self._client().put_object(
-            Bucket=settings.S3_BUCKET,
-            Key=key,
-            Body=blob,
-            ServerSideEncryption="AES256",  # SSE-S3 at rest
-            Metadata={"checksum": checksum},
-        )
-        return key
-
-    async def download(self, user_id: str, s3_key: str) -> bytes:
-        """Retrieve the blob stored at *s3_key*.
-
-        *user_id* is retained in the signature so higher-level code can
-        enforce ownership without re-parsing the key.
-
-        Raises:
-            ``botocore.exceptions.ClientError`` with code ``NoSuchKey`` if the
-            object does not exist.
-        """
-        response = self._client().get_object(
-            Bucket=settings.S3_BUCKET,
-            Key=s3_key,
-        )
-        return response["Body"].read()
-
-    async def delete(self, user_id: str, s3_key: str) -> None:
-        """Delete the object at *s3_key*.
-
-        S3 ``delete_object`` is idempotent — it succeeds even if the key does
-        not exist.
-        """
-        self._client().delete_object(
-            Bucket=settings.S3_BUCKET,
-            Key=s3_key,
-        )
-
-    async def list_keys(self, user_id: str, table: str) -> list[str]:
-        """Return all S3 keys for a given user + table combination.
-
-        Uses the prefix ``{user_id}/{table}/`` to scope the listing.
-        """
-        prefix = f"{user_id}/{table}/"
-        response = self._client().list_objects_v2(
-            Bucket=settings.S3_BUCKET,
-            Prefix=prefix,
-        )
-        return [obj["Key"] for obj in response.get("Contents", [])]

app/storage/encryption.py

@@ -1,32 +0,0 @@
-"""Integrity verification only — the backend NEVER decrypts user data."""
-
-from __future__ import annotations
-
-import hashlib
-import hmac
-
-from fastapi import HTTPException
-
-
-def verify_checksum(blob: bytes, checksum: str) -> bool:
-    """Return ``True`` if SHA-256(blob) matches *checksum*.
-
-    Uses ``hmac.compare_digest`` for constant-time comparison to prevent
-    timing-based side-channel attacks.
-    """
-    computed = hashlib.sha256(blob).hexdigest()
-    return hmac.compare_digest(computed, checksum)
-
-
-def reject_if_tampered(blob: bytes, checksum: str) -> None:
-    """Raise ``HTTP 400`` if the blob does not match its checksum.
-
-    Call this before storing or forwarding any client-provided blob.
-    The backend never holds decryption keys — this check only verifies
-    that the opaque bytes arrived intact.
-    """
-    if not verify_checksum(blob, checksum):
-        raise HTTPException(
-            status_code=400,
-            detail="Checksum mismatch: blob integrity check failed",
-        )

app/storage/vector_store.py

@@ -1,205 +0,0 @@
-"""Cloud vector store — wraps Pinecone (default) or Qdrant.
-
-Vectors are pre-encrypted blobs from the client.  The backend stores them
-alongside a deterministic 32-dim float representation derived from the blob's
-SHA-256 hash.  Semantic ANN search is not meaningful on encrypted data — this
-is a known trade-off documented in the backend plan.
-
-Isolation: Pinecone uses ``namespace=user_id``; Qdrant filters by
-``user_id`` payload field on a shared collection.
-"""
-
-from __future__ import annotations
-
-import base64
-import hashlib
-from typing import Any
-
-from pinecone import Pinecone
-from qdrant_client import QdrantClient
-from qdrant_client.models import FieldCondition, Filter, MatchValue, PointIdsList, PointStruct
-
-from app.config.settings import settings
-from app.schemas import VectorItem, VectorSearchResult
-
-_QDRANT_COLLECTION = "adiuva_vectors"
-
-
-def _blob_to_vector(blob: bytes) -> list[float]:
-    """Derive a 32-dim float vector from *blob* for storage purposes only.
-
-    Uses SHA-256 to produce a deterministic 32-byte fingerprint, then
-    normalises each byte to the range [-1.0, 1.0].  This vector carries no
-    semantic meaning on encrypted data.
-    """
-    return [(b - 128) / 128.0 for b in hashlib.sha256(blob).digest()]
-
-
-class VectorStore:
-    """Thin wrapper around Pinecone or Qdrant.
-
-    The backend to use is selected at runtime:
-    - Pinecone: when ``settings.PINECONE_API_KEY`` is non-empty.
-    - Qdrant: otherwise (requires ``settings.QDRANT_URL``).
-    """
-
-    def _use_pinecone(self) -> bool:
-        return bool(settings.PINECONE_API_KEY)
-
-    # ── Pinecone helpers ──────────────────────────────────────────────
-
-    def _pinecone_index(self) -> Any:
-        pc = Pinecone(api_key=settings.PINECONE_API_KEY)
-        return pc.Index(settings.PINECONE_INDEX)
-
-    # ── Qdrant helpers ────────────────────────────────────────────────
-
-    def _qdrant_client(self) -> Any:
-        return QdrantClient(
-            url=settings.QDRANT_URL,
-            api_key=settings.QDRANT_API_KEY or None,
-        )
-
-    # ── Public API ────────────────────────────────────────────────────
-
-    async def upsert(self, user_id: str, vectors: list[VectorItem]) -> None:
-        """Store encrypted vectors in the backend.
-
-        Each ``VectorItem.blob`` is base64-encoded and kept in metadata/payload
-        so it can be returned verbatim during search.
-
-        Args:
-            user_id: Used as Pinecone namespace or Qdrant payload field.
-            vectors: List of encrypted vector items from the client.
-        """
-        if self._use_pinecone():
-            await self._pinecone_upsert(user_id, vectors)
-        else:
-            await self._qdrant_upsert(user_id, vectors)
-
-    async def search(
-        self,
-        user_id: str,
-        query_blob: bytes,
-        top_k: int,
-    ) -> list[VectorSearchResult]:
-        """Query the vector store and return encrypted result blobs.
-
-        The query vector is derived from *query_blob* using the same
-        deterministic mapping as upsert.
-
-        Args:
-            user_id:    Scopes the search to this user's namespace.
-            query_blob: Encrypted query from the client.
-            top_k:      Maximum number of results to return.
-
-        Returns:
-            List of ``VectorSearchResult`` with ``id``, ``score``, and ``blob``.
-        """
-        if self._use_pinecone():
-            return await self._pinecone_search(user_id, query_blob, top_k)
-        return await self._qdrant_search(user_id, query_blob, top_k)
-
-    async def delete(self, user_id: str, vector_ids: list[str]) -> None:
-        """Remove vectors by ID, scoped to *user_id*.
-
-        Args:
-            user_id:    Namespace / payload filter to prevent cross-user deletion.
-            vector_ids: List of vector IDs to remove.
-        """
-        if self._use_pinecone():
-            await self._pinecone_delete(user_id, vector_ids)
-        else:
-            await self._qdrant_delete(user_id, vector_ids)
-
-    # ── Pinecone implementation ───────────────────────────────────────
-
-    async def _pinecone_upsert(self, user_id: str, vectors: list[VectorItem]) -> None:
-        index = self._pinecone_index()
-        records = [
-            {
-                "id": v.id,
-                "values": _blob_to_vector(v.blob),
-                "metadata": {
-                    "blob": base64.b64encode(v.blob).decode(),
-                    "checksum": v.checksum,
-                    "user_id": user_id,
-                },
-            }
-            for v in vectors
-        ]
-        index.upsert(vectors=records, namespace=user_id)
-
-    async def _pinecone_search(
-        self, user_id: str, query_blob: bytes, top_k: int
-    ) -> list[VectorSearchResult]:
-        index = self._pinecone_index()
-        query_vector = _blob_to_vector(query_blob)
-        response = index.query(
-            vector=query_vector,
-            top_k=top_k,
-            namespace=user_id,
-            include_metadata=True,
-        )
-        results: list[VectorSearchResult] = []
-        for match in response.get("matches", []):
-            blob_bytes = base64.b64decode(match["metadata"]["blob"])
-            results.append(
-                VectorSearchResult(
-                    id=match["id"],
-                    score=match["score"],
-                    blob=blob_bytes,
-                )
-            )
-        return results
-
-    async def _pinecone_delete(self, user_id: str, vector_ids: list[str]) -> None:
-        index = self._pinecone_index()
-        index.delete(ids=vector_ids, namespace=user_id)
-
-    # ── Qdrant implementation ─────────────────────────────────────────
-
-    async def _qdrant_upsert(self, user_id: str, vectors: list[VectorItem]) -> None:
-        client = self._qdrant_client()
-        points = [
-            PointStruct(
-                id=v.id,
-                vector=_blob_to_vector(v.blob),
-                payload={
-                    "blob": base64.b64encode(v.blob).decode(),
-                    "checksum": v.checksum,
-                    "user_id": user_id,
-                },
-            )
-            for v in vectors
-        ]
-        client.upsert(collection_name=_QDRANT_COLLECTION, points=points)
-
-    async def _qdrant_search(
-        self, user_id: str, query_blob: bytes, top_k: int
-    ) -> list[VectorSearchResult]:
-        client = self._qdrant_client()
-        query_vector = _blob_to_vector(query_blob)
-        hits = client.search(
-            collection_name=_QDRANT_COLLECTION,
-            query_vector=query_vector,
-            query_filter=Filter(
-                must=[FieldCondition(key="user_id", match=MatchValue(value=user_id))]
-            ),
-            limit=top_k,
-        )
-        return [
-            VectorSearchResult(
-                id=str(hit.id),
-                score=hit.score,
-                blob=base64.b64decode(hit.payload["blob"]),
-            )
-            for hit in hits
-        ]
-
-    async def _qdrant_delete(self, user_id: str, vector_ids: list[str]) -> None:
-        client = self._qdrant_client()
-        client.delete(
-            collection_name=_QDRANT_COLLECTION,
-            points_selector=PointIdsList(points=vector_ids),
-        )

docker-compose.yml

@@ -7,7 +7,7 @@ services:
       - path: .env
         required: false
     environment:
-      DATABASE_URL: postgresql+asyncpg://postgres:postgres@db:5432/adiuva
+      DATABASE_URL: postgresql+asyncpg://postgres:postgres@db:5432/adiuvai
       GITHUB_COPILOT_TOKEN_DIR: /root/.config/litellm/github_copilot
     volumes:
       - copilot_tokens:/root/.config/litellm/github_copilot
@@ -21,7 +21,7 @@ services:
     environment:
       POSTGRES_USER: postgres
       POSTGRES_PASSWORD: postgres
-      POSTGRES_DB: adiuva
+      POSTGRES_DB: adiuvai
     volumes:
       - postgres_data:/var/lib/postgresql/data
     healthcheck:
@@ -36,37 +36,6 @@ services:
   #   image: redis:7-alpine
   #   restart: unless-stopped
 
-  # ── Local S3-compatible storage (MinIO) ──
-  minio:
-    image: minio/minio:latest
-    command: server /data --console-address ":9001"
-    ports:
-      - "9000:9000"
-      - "9001:9001"
-    environment:
-      MINIO_ROOT_USER: minioadmin
-      MINIO_ROOT_PASSWORD: minioadmin
-    volumes:
-      - minio_data:/data
-    healthcheck:
-      test: ["CMD", "mc", "ready", "local"]
-      interval: 5s
-      timeout: 5s
-      retries: 5
-    restart: unless-stopped
-
-  # ── Local vector store (Qdrant) ──
-  qdrant:
-    image: qdrant/qdrant:latest
-    ports:
-      - "6333:6333"
-      - "6334:6334"
-    volumes:
-      - qdrant_data:/qdrant/storage
-    restart: unless-stopped
-
 volumes:
   postgres_data:
-  minio_data:
-  qdrant_data:
   copilot_tokens:

docs/MICROSERVICES_ARCHITECTURE.md

@@ -1,941 +0,0 @@
-# Adiuva — Architettura Microservizi (MVP)
-
-## Panoramica
-
-Il monolite viene suddiviso in **4 servizi MVP** + un **API Gateway (Traefik)**, orchestrati con Docker Compose su un singolo VPS raggiungibile via Cloudflare.
-
-> **Fuori dall'MVP**: Storage Service (S3/backup CRUD) e Plugin Service (marketplace). Verranno aggiunti come servizi indipendenti in una fase successiva.
-
-```
-                          ┌──────────────┐
-                          │  Cloudflare  │
-                          │  (DNS + CDN) │
-                          └──────┬───────┘
-                                 │ HTTPS / WSS
-                          ┌──────▼───────┐
-                          │   Traefik    │
-                          │ API Gateway  │
-                          │  (routing,   │
-                          │   TLS, rate  │
-                          │   limiting)  │
-                          └──────┬───────┘
-                                 │
-          ┌──────────┬───────────┼───────────┐
-          │          │           │           │
-    ┌─────▼────┐ ┌───▼───┐ ┌────▼────┐ ┌────▼───┐
-    │  Auth    │ │  Chat │ │  Agent  │ │Billing │
-    │ Service  │ │Service│ │ Service │ │Service │
-    └─────┬────┘ └───┬───┘ └────┬────┘ └────┬───┘
-          │          │          │           │
-    ┌─────▼──────────▼──────────▼───────────▼────┐
-    │              Infrastruttura                 │
-    │  PostgreSQL  │  Redis  │  Qdrant            │
-    └─────────────────────────────────────────────┘
-```
-
----
-
-## 1. Suddivisione dei Servizi
-
-### 1.1 Auth Service (`auth-service`)
-
-**Responsabilità**: Registrazione, login, refresh token, profilo utente, encryption key.
-
-| Endpoint originale | Metodo |
-|---|---|
-| `/api/v1/auth/register` | POST |
-| `/api/v1/auth/login` | POST |
-| `/api/v1/auth/refresh` | POST |
-| `/api/v1/auth/me` | GET / PUT |
-
-**Database**: Tabelle `users`, `refresh_tokens` (PostgreSQL condiviso, schema `auth`).
-
-**Modifica chiave — JWT con RS256**:
-Il monolite usa un `SECRET_KEY` simmetrico (HS256). Con i microservizi, passare a **RS256** (asimmetrico):
-- L'Auth Service firma i JWT con la **chiave privata**.
-- Tutti gli altri servizi verificano i JWT con la **chiave pubblica** senza mai contattare l'Auth Service.
-- La chiave pubblica viene esposta via `GET /api/v1/auth/.well-known/jwks.json` oppure montata come volume condiviso.
-
-```python
-# auth-service/app/auth/jwt.py
-from cryptography.hazmat.primitives.asymmetric import rsa
-from jose import jwt
-
-PRIVATE_KEY = ...  # Da env/secret
-PUBLIC_KEY = ...   # Derivata o da env
-
-def create_access_token(user_id: str, tier: str) -> str:
-    return jwt.encode(
-        {"sub": user_id, "tier": tier, "exp": ...},
-        PRIVATE_KEY,
-        algorithm="RS256",
-    )
-```
-
-```python
-# shared/auth.py  (usato da tutti gli altri servizi)
-from jose import jwt
-
-PUBLIC_KEY = ...  # Volume montato o fetched da JWKS endpoint
-
-def verify_token(token: str) -> dict:
-    return jwt.decode(token, PUBLIC_KEY, algorithms=["RS256"])
-```
-
-**Scaling**: 2 repliche sufficienti, stateless. Rate-limit dedicato su `/login` e `/register`.
-
----
-
-### 1.2 Chat Service (`chat-service`) ⭐ Real-time
-
-**Responsabilità**: WebSocket device connection, home chat, floating chat, memory middleware, streaming LLM responses verso il client.
-
-Questo servizio gestisce la **connessione persistente** con l'app Electron e le interazioni **real-time** dell'utente (chat home, floating chat). È il proprietario della WebSocket.
-
-| Endpoint | Tipo |
-|---|---|
-| `/api/v1/ws/device` | WebSocket (connessione persistente) |
-| `/api/v1/chat` | POST (REST fallback) |
-
-**Moduli inclusi**: `deep_agent`, `memory_middleware`, `ws_context`, `device_manager` (Redis-backed), `output_formatter`, `llm`, tutti gli agent tools (`task_agent`, `project_agent`, `note_agent`, `timeline_agent`).
-
-**Perché separato dall'Agent Service**: Il Chat Service tiene la WebSocket aperta e risponde in tempo reale (streaming). Scalare aggiungendo repliche è semplice con sticky sessions + Redis pub/sub per il cross-instance routing dei tool_call.
-
-**Scaling**: 2–N repliche. Sticky cookies per le WS + Redis per cross-instance.
-
----
-
-### 1.3 Agent Service (`agent-service`) ⭐ Batch
-
-**Responsabilità**: Batch agent processing (directory scanning, file classification, entity extraction), agent setup journeys, agent configuration CRUD.
-
-Questo servizio gestisce i processi **long-running** e **CPU-intensive**: scansione filesystem, classificazione file con LLM, estrazione entità in batch. Non possiede la WebSocket — comunica con il device dell'utente tramite **Redis pub/sub** passando per il Chat Service.
-
-| Endpoint | Tipo |
-|---|---|
-| `/api/v1/agents/catalog` | GET |
-| `/api/v1/agents/can-create` | POST |
-| `/api/v1/agents/trigger` | POST |
-| `/api/v1/agents/journey/start` | POST (o WS relay) |
-| `/api/v1/agents/journey/message` | POST (o WS relay) |
-
-**Moduli inclusi**: `agent_runner`, `agent_registry`, `filesystem_agent`, `llm`.
-
-**Flusso tool-call cross-service** (l'Agent Service non ha la WS):
-
-```
-┌──────────────┐            ┌──────────────┐            ┌──────────┐
-│ Agent Service│            │    Redis     │            │  Chat    │
-│ (batch run)  │            │              │            │ Service  │
-│              │            │              │            │ (ha WS)  │
-│ 1. Needs to  │  PUBLISH   │              │ SUBSCRIBE  │          │
-│    read file ├───────────►│tool_call:u123├───────────►│ 2. Invia │
-│    from      │            │              │            │    al    │
-│    device    │            │              │            │    device│
-│              │            │              │            │    via WS│
-│              │  SUBSCRIBE │              │  PUBLISH   │          │
-│ 4. Riceve   ◄────────────┤tool_result:id│◄───────────┤ 3. Device│
-│    risultato │            │              │            │    reply │
-└──────────────┘            └──────────────┘            └──────────┘
-```
-
-**Scaling**: 1–N repliche. Completamente stateless, scala indipendentemente dalla chat. Ogni replica processa batch job diversi. Può essere scalato a 0 se non ci sono agent attivi (risparmio risorse).
-
-**Vantaggio dello split**: Se 50 utenti triggerano agenti batch contemporaneamente, il Chat Service non ne risente — le risposte real-time rimangono veloci.
-
----
-
-### 1.4 Billing Service (`billing-service`)
-
-**Responsabilità**: Stripe checkout, webhook, subscription management.
-
-| Endpoint originale | Metodo |
-|---|---|
-| `/api/v1/billing/checkout` | POST |
-| `/api/v1/billing/webhook` | POST |
-| `/api/v1/billing/subscription` | GET / DELETE |
-
-**Database**: Tabelle `subscriptions` (schema `billing`).
-
-**Comunicazione inter-servizio**: Quando Stripe invia un webhook e il tier cambia, il Billing Service pubblica un evento su **Redis pub/sub** channel `tier_changed:{user_id}`. L'Auth Service aggiorna il campo `tier` nella tabella users. Al prossimo token refresh il JWT conterrà il tier aggiornato.
-
-**Scaling**: 1 replica sufficiente. Basso traffico.
-
----
-
-### 1.5 Servizi esclusi dall'MVP
-
-I seguenti servizi verranno aggiunti post-MVP come servizi indipendenti:
-
-| Servizio | Responsabilità | Note |
-|---|---|---|
-| **Storage Service** | S3 blobs CRUD, vector ops, backup | Le funzionalità vector/embed possono restare nel Chat Service per il MVP |
-| **Plugin Service** | Marketplace, install, revenue split | Feature non critica per il lancio |
-
----
-
-## 2. Tier Check — Dove e Come
-
-Il tier dell'utente (free/pro/power/team) determina rate-limiting, quote e accesso a funzionalità. Con i microservizi, **ogni servizio controlla il tier autonomamente** senza chiamare l'Auth Service.
-
-### Strategia: Tier nel JWT
-
-L'Auth Service include il `tier` come claim nel JWT al momento del login/refresh:
-
-```json
-{
-  "sub": "user_123",
-  "tier": "pro",
-  "exp": 1742515200,
-  "iat": 1742511600
-}
-```
-
-Ogni servizio:
-1. Decodifica il JWT con la chiave pubblica (già lo fa per l'auth)
-2. Legge `payload["tier"]` — **zero chiamate extra**
-3. Applica le sue regole di enforcement localmente
-
-```python
-# shared/auth.py — dependency FastAPI condivisa
-from fastapi import Depends, HTTPException, Request
-from jose import jwt
-
-PUBLIC_KEY = ...
-
-class CurrentUser:
-    def __init__(self, user_id: str, tier: str):
-        self.user_id = user_id
-        self.tier = tier
-
-async def get_current_user(request: Request) -> CurrentUser:
-    token = request.headers.get("Authorization", "").removeprefix("Bearer ")
-    payload = jwt.decode(token, PUBLIC_KEY, algorithms=["RS256"])
-    return CurrentUser(user_id=payload["sub"], tier=payload["tier"])
-
-def require_tier(*allowed_tiers: str):
-    """Dependency che blocca se il tier non è tra quelli ammessi."""
-    async def check(user: CurrentUser = Depends(get_current_user)):
-        if user.tier not in allowed_tiers:
-            raise HTTPException(403, "Tier insufficient")
-        return user
-    return check
-```
-
-### Cosa succede quando il tier cambia (upgrade/downgrade)?
-
-```
-┌──────────┐  Stripe webhook   ┌──────────┐  tier_changed   ┌──────────┐
-│  Stripe  │ ─────────────────►│ Billing  │ ───────────────►│   Auth   │
-│          │                    │ Service  │  (Redis pub/sub) │ Service  │
-└──────────┘                    └──────────┘                  └────┬─────┘
-                                                                   │
-                                                          UPDATE users
-                                                          SET tier = 'power'
-                                                                   │
-                                                    Al prossimo /refresh
-                                                    il JWT conterrà tier='power'
-```
-
-**Latenza del cambio**: Il tier si propaga al prossimo token refresh (tipicamente 15–30 min, o il client può forzare un refresh immediato dopo il checkout). Per il billing webhook, il downgrade può essere forzato invalidando il refresh token su Redis → il client è obbligato a ri-autenticarsi.
-
-### Dove si applica in ciascun servizio
-
-| Servizio | Enforcement |
-|---|---|
-| **Auth Service** | Nessuno (è lui che scrive il tier) |
-| **Chat Service** | Rate-limit per tier (req/min), quota messaggi |
-| **Agent Service** | Max agent configs, max runs/day, max concurrent batches |
-| **Billing Service** | Nessuno (gestisce i tier, non li consuma) |
-
-### Rate-limit distribuito via Redis
-
-Poiché ogni servizio ha le sue repliche, il rate-limiting deve essere **condiviso** via Redis:
-
-```python
-# shared/middleware/rate_limit.py
-import redis.asyncio as aioredis
-
-class DistributedRateLimiter:
-    def __init__(self, redis: aioredis.Redis):
-        self._redis = redis
-
-    async def check(self, user_id: str, tier: str, service: str) -> bool:
-        limits = {"free": 20, "pro": 60, "power": 120, "team": 200}
-        max_req = limits.get(tier, 20)
-        key = f"rate:{service}:{user_id}"
-
-        pipe = self._redis.pipeline()
-        pipe.incr(key)
-        pipe.expire(key, 60)
-        count, _ = await pipe.execute()
-
-        return count <= max_req
-```
-
----
-
-## 3. WebSocket con Scaling Orizzontale — Il Problema Chiave
-
-`DeviceConnectionManager` è un **singleton in-memory**:
-
-```python
-class DeviceConnectionManager:
-    def __init__(self):
-        self._connections: dict[str, DeviceConnection] = {}  # ← In-memory!
-```
-
-Con N istanze del Chat Service, il device si connette a **una sola** istanza. Quando un'altra istanza deve inviare un `tool_call` a quel device (es. un agent trigger da un'API call), non trova la connessione.
-
-### La soluzione: Redis Pub/Sub + Registry
-
-```
-┌──────────────────────────────────────────────────────────────┐
-│                     Redis                                    │
-│                                                              │
-│  Hash: ws:connections                                        │
-│    user_123 → instance_A                                     │
-│    user_456 → instance_B                                     │
-│                                                              │
-│  Pub/Sub channels:                                           │
-│    tool_call:{user_id}  → tool call payloads                 │
-│    tool_result:{call_id} → tool result payloads              │
-│    stream:{user_id}     → text_chunk streaming               │
-└──────────────────────────────────────────────────────────────┘
-
- Instance A (ha WS di user_123)     Instance B (deve chiamare tool su user_123)
- ┌───────────────────────┐          ┌───────────────────────┐
- │  1. Sottoscrive a     │          │  1. Lookup Redis Hash │
- │     tool_call:user_123│          │     → user_123 è su A │
- │                       │          │                       │
- │  2. Riceve tool_call  │◄─────────│  2. PUBLISH           │
- │     da Redis channel  │          │    tool_call:user_123 │
- │                       │          │    {id, action, ...}  │
- │  3. Invia al device   │          │                       │
- │     via WS            │          │  4. SUBSCRIBE         │
- │                       │          │    tool_result:{id}   │
- │  4. Device risponde   │          │                       │
- │     tool_result       │──────────│► 5. Riceve risultato  │
- │                       │          │                       │
- │  5. PUBLISH           │          │                       │
- │    tool_result:{id}   │          │                       │
- └───────────────────────┘          └───────────────────────┘
-```
-
-### Implementazione: `RedisDeviceManager`
-
-```python
-# chat-service/app/core/device_manager.py
-
-import asyncio
-import json
-import os
-import redis.asyncio as aioredis
-from dataclasses import dataclass, field
-from fastapi import WebSocket
-
-INSTANCE_ID = os.environ.get("INSTANCE_ID", os.urandom(8).hex())
-
-@dataclass
-class LocalConnection:
-    ws: WebSocket
-    device_id: str
-    pending_calls: dict[str, asyncio.Future[dict]] = field(default_factory=dict)
-
-
-class RedisDeviceManager:
-    """Device manager backed by Redis for cross-instance communication."""
-
-    def __init__(self, redis_url: str = "redis://redis:6379"):
-        self._redis = aioredis.from_url(redis_url)
-        self._pubsub = self._redis.pubsub()
-        self._local: dict[str, LocalConnection] = {}  # Solo connessioni locali
-        self._remote_futures: dict[str, asyncio.Future[dict]] = {}
-
-    async def start(self):
-        """Avvia il listener Redis per tool_call in arrivo."""
-        asyncio.create_task(self._listen_tool_calls())
-
-    # ── Registrazione ──
-
-    async def register(self, user_id: str, device_id: str, ws: WebSocket):
-        # Registra localmente
-        self._local[user_id] = LocalConnection(ws=ws, device_id=device_id)
-        # Registra in Redis quale istanza ha la connessione
-        await self._redis.hset("ws:connections", user_id, INSTANCE_ID)
-        # Sottoscrivi ai tool_call per questo utente
-        await self._pubsub.subscribe(f"tool_call:{user_id}")
-
-    async def unregister(self, user_id: str):
-        conn = self._local.pop(user_id, None)
-        if conn:
-            for fut in conn.pending_calls.values():
-                if not fut.done():
-                    fut.cancel()
-        await self._redis.hdel("ws:connections", user_id)
-        await self._pubsub.unsubscribe(f"tool_call:{user_id}")
-
-    # ── Presenza ──
-
-    async def is_online(self, user_id: str) -> bool:
-        return await self._redis.hexists("ws:connections", user_id)
-
-    # ── Tool-call round-trip (cross-instance) ──
-
-    async def execute_tool_call(self, user_id: str, payload: dict) -> dict:
-        """
-        Invia un tool_call al device dell'utente.
-        Funziona sia che la WS sia locale che su un'altra istanza.
-        """
-        call_id = payload["id"]
-
-        # Caso 1: connessione locale → invio diretto
-        if user_id in self._local:
-            conn = self._local[user_id]
-            loop = asyncio.get_event_loop()
-            fut: asyncio.Future[dict] = loop.create_future()
-            conn.pending_calls[call_id] = fut
-            await conn.ws.send_text(json.dumps({"type": "tool_call", **payload}))
-            return await asyncio.wait_for(fut, timeout=30.0)
-
-        # Caso 2: connessione remota → Redis pub/sub
-        loop = asyncio.get_event_loop()
-        fut = loop.create_future()
-        self._remote_futures[call_id] = fut
-
-        # Sottoscrivi al canale di risposta
-        result_channel = f"tool_result:{call_id}"
-        await self._pubsub.subscribe(result_channel)
-
-        # Pubblica il tool_call
-        await self._redis.publish(
-            f"tool_call:{user_id}",
-            json.dumps(payload),
-        )
-
-        try:
-            return await asyncio.wait_for(fut, timeout=30.0)
-        finally:
-            self._remote_futures.pop(call_id, None)
-            await self._pubsub.unsubscribe(result_channel)
-
-    # ── Risoluzione tool_result (da WS locale) ──
-
-    def resolve_local(self, user_id: str, call_id: str, result: dict):
-        conn = self._local.get(user_id)
-        if conn:
-            fut = conn.pending_calls.pop(call_id, None)
-            if fut and not fut.done():
-                fut.set_result(result)
-
-    async def resolve_and_publish(self, user_id: str, call_id: str, result: dict):
-        """Chiamato quando il device locale invia un tool_result."""
-        self.resolve_local(user_id, call_id, result)
-        # Pubblica anche su Redis per l'istanza remota che aspetta
-        await self._redis.publish(
-            f"tool_result:{call_id}",
-            json.dumps(result),
-        )
-
-    # ── Listener Redis ──
-
-    async def _listen_tool_calls(self):
-        """Loop che ascolta i tool_call in arrivo da altre istanze."""
-        async for message in self._pubsub.listen():
-            if message["type"] != "message":
-                continue
-            channel = message["channel"]
-            if isinstance(channel, bytes):
-                channel = channel.decode()
-
-            data = json.loads(message["data"])
-
-            if channel.startswith("tool_call:"):
-                # Un'altra istanza vuole che inviamo un tool_call al nostro device
-                user_id = channel.split(":", 1)[1]
-                conn = self._local.get(user_id)
-                if conn:
-                    await conn.ws.send_text(json.dumps({"type": "tool_call", **data}))
-
-            elif channel.startswith("tool_result:"):
-                # Risposta a un tool_call che abbiamo inviato tramite Redis
-                call_id = channel.split(":", 1)[1]
-                fut = self._remote_futures.pop(call_id, None)
-                if fut and not fut.done():
-                    fut.set_result(data)
-
-    # ── Stream cross-instance ──
-
-    async def publish_stream_chunk(self, user_id: str, chunk: dict):
-        """Pubblica un chunk di streaming su Redis (per REST→WS relay)."""
-        await self._redis.publish(f"stream:{user_id}", json.dumps(chunk))
-```
-
----
-
-## 4. Struttura Directory Proposta (MVP)
-
-```
-adiuva-api/
-├── docker-compose.yml          # Orchestrazione completa
-├── docker-compose.dev.yml      # Override per sviluppo locale
-├── shared/                     # Codice condiviso (montato come volume)
-│   ├── auth.py                 # JWT verification (chiave pubblica)
-│   ├── schemas.py              # Pydantic schemas condivisi
-│   ├── middleware/
-│   │   ├── rate_limit.py       # DistributedRateLimiter (Redis)
-│   │   └── sanitizer.py
-│   └── models/
-│       └── base.py             # SQLAlchemy base condivisa
-│
-├── auth-service/
-│   ├── Dockerfile
-│   ├── requirements.txt
-│   └── app/
-│       ├── main.py
-│       ├── config.py
-│       ├── db.py
-│       ├── models.py           # users, refresh_tokens
-│       ├── routes/
-│       │   └── auth.py
-│       └── services/
-│           ├── jwt_service.py  # RS256 signing
-│           └── user_service.py
-│
-├── chat-service/
-│   ├── Dockerfile
-│   ├── requirements.txt
-│   └── app/
-│       ├── main.py
-│       ├── config.py
-│       ├── db.py
-│       ├── models.py           # memory_*
-│       ├── routes/
-│       │   ├── device_ws.py    # WS connection owner
-│       │   └── chat.py         # REST fallback
-│       ├── core/
-│       │   ├── device_manager.py   # RedisDeviceManager
-│       │   ├── deep_agent.py       # Home + floating chat
-│       │   ├── memory_middleware.py
-│       │   ├── ws_context.py
-│       │   ├── output_formatter.py
-│       │   └── llm.py
-│       └── agents/                 # Tool definitions (used by deep_agent)
-│           ├── task_agent.py
-│           ├── project_agent.py
-│           ├── note_agent.py
-│           └── timeline_agent.py
-│
-├── agent-service/
-│   ├── Dockerfile
-│   ├── requirements.txt
-│   └── app/
-│       ├── main.py
-│       ├── config.py
-│       ├── db.py
-│       ├── models.py           # agent_run_logs, local/cloud_agent_configs
-│       ├── routes/
-│       │   ├── agents.py       # catalog, can-create, trigger
-│       │   └── agent_setup.py  # journey start/message
-│       ├── core/
-│       │   ├── agent_runner.py     # Batch classify → process
-│       │   ├── agent_registry.py
-│       │   ├── redis_executor.py   # execute_on_client via Redis pub/sub
-│       │   └── llm.py
-│       └── agents/
-│           ├── task_agent.py       # Tool definitions (batch context)
-│           ├── project_agent.py
-│           ├── note_agent.py
-│           ├── timeline_agent.py
-│           └── filesystem_agent.py
-│
-├── billing-service/
-│   ├── Dockerfile
-│   ├── requirements.txt
-│   └── app/
-│       ├── main.py
-│       ├── config.py
-│       ├── db.py
-│       ├── models.py           # subscriptions
-│       ├── routes/
-│       │   └── billing.py
-│       └── services/
-│           ├── stripe_service.py
-│           └── tier_manager.py
-│
-└── infra/
-    ├── traefik/
-    │   └── traefik.yml
-    ├── keys/
-    │   ├── jwt_private.pem     # Solo auth-service
-    │   └── jwt_public.pem      # Tutti i servizi
-    └── alembic/                # Migrazioni condivise o per-servizio
-```
-
----
-
-## 5. Docker Compose — Configurazione MVP
-
-```yaml
-# docker-compose.yml
-
-services:
-
-  # ══════════════════════════════════════════════════════════
-  # API Gateway
-  # ══════════════════════════════════════════════════════════
-  traefik:
-    image: traefik:v3.2
-    command:
-      - "--api.insecure=true"
-      - "--providers.docker=true"
-      - "--providers.docker.exposedbydefault=false"
-      - "--entrypoints.web.address=:80"
-      - "--entrypoints.websecure.address=:443"
-      - "--entrypoints.web.http.redirections.entrypoint.to=websecure"
-    ports:
-      - "80:80"
-      - "443:443"
-      - "8080:8080"   # Dashboard Traefik (disabilitare in prod)
-    volumes:
-      - /var/run/docker.sock:/var/run/docker.sock:ro
-      - ./infra/certs:/certs:ro
-    restart: unless-stopped
-
-  # ══════════════════════════════════════════════════════════
-  # Auth Service (2 repliche)
-  # ══════════════════════════════════════════════════════════
-  auth-service:
-    build: ./auth-service
-    deploy:
-      replicas: 2
-    env_file: .env
-    environment:
-      DATABASE_URL: postgresql+asyncpg://postgres:postgres@db:5432/adiuva
-      REDIS_URL: redis://redis:6379
-      JWT_PRIVATE_KEY_FILE: /run/secrets/jwt_private_key
-      SERVICE_NAME: auth
-    secrets:
-      - jwt_private_key
-      - jwt_public_key
-    labels:
-      - "traefik.enable=true"
-      - "traefik.http.routers.auth.rule=PathPrefix(`/api/v1/auth`)"
-      - "traefik.http.services.auth.loadbalancer.server.port=8000"
-    depends_on:
-      db:
-        condition: service_healthy
-      redis:
-        condition: service_healthy
-
-  # ══════════════════════════════════════════════════════════
-  # Chat Service — Real-time WS + Chat (scalabile)
-  # ══════════════════════════════════════════════════════════
-  chat-service:
-    build: ./chat-service
-    deploy:
-      replicas: 2
-    env_file: .env
-    environment:
-      DATABASE_URL: postgresql+asyncpg://postgres:postgres@db:5432/adiuva
-      REDIS_URL: redis://redis:6379
-      JWT_PUBLIC_KEY_FILE: /run/secrets/jwt_public_key
-      SERVICE_NAME: chat
-    secrets:
-      - jwt_public_key
-    labels:
-      - "traefik.enable=true"
-      # REST chat endpoint
-      - "traefik.http.routers.chat.rule=PathPrefix(`/api/v1/chat`)"
-      - "traefik.http.services.chat.loadbalancer.server.port=8000"
-      # WebSocket route con sticky session
-      - "traefik.http.routers.ws.rule=PathPrefix(`/api/v1/ws`)"
-      - "traefik.http.routers.ws.service=chat-ws"
-      - "traefik.http.services.chat-ws.loadbalancer.server.port=8000"
-      - "traefik.http.services.chat-ws.loadbalancer.sticky.cookie.name=ws_affinity"
-      - "traefik.http.services.chat-ws.loadbalancer.sticky.cookie.httpOnly=true"
-    depends_on:
-      db:
-        condition: service_healthy
-      redis:
-        condition: service_healthy
-
-  # ══════════════════════════════════════════════════════════
-  # Agent Service — Batch processing (scalabile indipendentemente)
-  # ══════════════════════════════════════════════════════════
-  agent-service:
-    build: ./agent-service
-    deploy:
-      replicas: 2
-    env_file: .env
-    environment:
-      DATABASE_URL: postgresql+asyncpg://postgres:postgres@db:5432/adiuva
-      REDIS_URL: redis://redis:6379
-      JWT_PUBLIC_KEY_FILE: /run/secrets/jwt_public_key
-      SERVICE_NAME: agent
-    secrets:
-      - jwt_public_key
-    labels:
-      - "traefik.enable=true"
-      - "traefik.http.routers.agents.rule=PathPrefix(`/api/v1/agents`)"
-      - "traefik.http.services.agents.loadbalancer.server.port=8000"
-    depends_on:
-      db:
-        condition: service_healthy
-      redis:
-        condition: service_healthy
-
-  # ══════════════════════════════════════════════════════════
-  # Billing Service (1 replica)
-  # ══════════════════════════════════════════════════════════
-  billing-service:
-    build: ./billing-service
-    deploy:
-      replicas: 1
-    env_file: .env
-    environment:
-      DATABASE_URL: postgresql+asyncpg://postgres:postgres@db:5432/adiuva
-      REDIS_URL: redis://redis:6379
-      JWT_PUBLIC_KEY_FILE: /run/secrets/jwt_public_key
-      SERVICE_NAME: billing
-    secrets:
-      - jwt_public_key
-    labels:
-      - "traefik.enable=true"
-      - "traefik.http.routers.billing.rule=PathPrefix(`/api/v1/billing`)"
-      - "traefik.http.services.billing.loadbalancer.server.port=8000"
-    depends_on:
-      db:
-        condition: service_healthy
-      redis:
-        condition: service_healthy
-
-  # ══════════════════════════════════════════════════════════
-  # Infrastruttura
-  # ══════════════════════════════════════════════════════════
-  db:
-    image: pgvector/pgvector:pg16
-    environment:
-      POSTGRES_USER: postgres
-      POSTGRES_PASSWORD: postgres
-      POSTGRES_DB: adiuva
-    volumes:
-      - postgres_data:/var/lib/postgresql/data
-    healthcheck:
-      test: ["CMD-SHELL", "pg_isready -U postgres"]
-      interval: 5s
-      timeout: 5s
-      retries: 5
-    restart: unless-stopped
-
-  redis:
-    image: redis:7-alpine
-    command: redis-server --maxmemory 256mb --maxmemory-policy allkeys-lru
-    volumes:
-      - redis_data:/data
-    healthcheck:
-      test: ["CMD", "redis-cli", "ping"]
-      interval: 5s
-      timeout: 3s
-      retries: 5
-    restart: unless-stopped
-
-  qdrant:
-    image: qdrant/qdrant:latest
-    volumes:
-      - qdrant_data:/qdrant/storage
-    restart: unless-stopped
-
-secrets:
-  jwt_private_key:
-    file: ./infra/keys/jwt_private.pem
-  jwt_public_key:
-    file: ./infra/keys/jwt_public.pem
-
-volumes:
-  postgres_data:
-  redis_data:
-  qdrant_data:
-```
-
----
-
-## 6. Configurazione Cloudflare + VPS
-
-### 6.1 DNS
-
-```
-api.tuodominio.com  →  A record  →  IP del VPS
-                    →  Proxy: ON (orange cloud)
-```
-
-### 6.2 Cloudflare Settings
-
-| Setting | Valore | Motivo |
-|---------|--------|--------|
-| SSL/TLS mode | **Full (Strict)** | Cloudflare ↔ VPS con certificato valido |
-| WebSocket | **ON** | Necessario per `/api/v1/ws/device` |
-| Proxy timeout | **100s** (Enterprise) o default | Le LLM calls possono durare 30s+ |
-| Under Attack Mode | Off (attivare se necessario) | |
-
-### 6.3 TLS sul VPS
-
-Due opzioni:
-- **Opzione A (consigliata)**: Cloudflare Origin Certificate → montato in Traefik
-- **Opzione B**: Let's Encrypt via Traefik (con DNS challenge Cloudflare)
-
-```yaml
-# traefik.yml — con Cloudflare Origin Certificate
-entryPoints:
-  websecure:
-    address: ":443"
-
-tls:
-  certificates:
-    - certFile: /certs/origin.pem
-      keyFile: /certs/origin-key.pem
-```
-
-### 6.4 Rete VPS
-
-```bash
-# UFW firewall — solo Cloudflare può raggiungere le porte 80/443
-# https://www.cloudflare.com/ips/
-ufw default deny incoming
-ufw allow from 173.245.48.0/20 to any port 443
-ufw allow from 103.21.244.0/22 to any port 443
-# ... (tutti gli IP range di Cloudflare)
-ufw allow ssh
-ufw enable
-```
-
----
-
-## 7. Comunicazione Inter-Servizio
-
-### 7.1 Redis Pub/Sub — Event Bus
-
-```
-┌──────────┐  tier_changed:user_123   ┌──────────┐
-│ Billing  │ ────────────────────────► │   Auth   │
-│ Service  │                           │ Service  │
-└──────────┘                           └──────────┘
-
-┌──────────┐  tool_call:user_123      ┌──────────┐
-│  Agent   │ ────────────────────────► │   Chat   │
-│ Service  │                           │ Service  │
-│ (batch)  │ ◄────────────────────────│ (ha WS)  │
-└──────────┘  tool_result:{call_id}    └──────────┘
-```
-
-### 7.2 Health Checks e Service Discovery
-
-Traefik gestisce automaticamente il service discovery via Docker labels. I servizi non devono conoscersi tra loro — comunicano solo via:
-- **Redis pub/sub** (tool-call cross-instance, tier events)
-- **Redis hash** (stato condiviso: `ws:connections`, rate-limit counters)
-- **PostgreSQL** (dati persistenti condivisi)
-
----
-
-## 8. Piano di Migrazione Incrementale (MVP)
-
-### Fase 1 — Preparazione (nel monolite attuale)
-1. Aggiungere Redis al `docker-compose.yml` attuale
-2. Migrare JWT da HS256 → RS256 (backward-compatible: accetta entrambi per un periodo)
-3. Implementare `RedisDeviceManager` come drop-in replacement del singleton in-memory
-4. Estrarre `shared/` con auth verification, schemas, middleware
-
-### Fase 2 — Auth Service (primo split)
-1. Estrarre `auth.py` routes + models in `auth-service/`
-2. Verificare che i JWT firmati da `auth-service` vengano validati dal monolite
-3. Aggiungere Traefik e routare `/api/v1/auth/*` al nuovo servizio
-4. Il monolite continua a servire tutto il resto
-
-### Fase 3 — Billing Service
-1. Estrarre billing routes, Stripe service, tier manager
-2. Configurare Redis pub/sub per `tier_changed` events
-3. Routare via Traefik
-
-### Fase 4 — Split Chat + Agent (il più delicato)
-1. Il monolite residuo contiene WS + chat + agents
-2. Separare Agent Service: estrarre `agent_runner`, `agent_registry`, `agent_setup`, route `/agents/*`
-3. Implementare `redis_executor.py` nell'Agent Service per tool-call via Redis
-4. Il Chat Service resta proprietario della WS e sottoscrive i canali `tool_call:{user_id}`
-5. Testare: trigger agent dall'Agent Service → tool_call via Redis → Chat Service → WS → device → risposta
-
-### Fase 5 — Scaling test
-1. Scalare Chat Service a 2 repliche, verificare sticky sessions
-2. Scalare Agent Service a 2 repliche, verificare batch processing distribuito
-3. Monitoring (Prometheus + Grafana) per ogni servizio
-
----
-
-## 9. Monitoraggio e Logging
-
-```yaml
-# Aggiungere al docker-compose.yml
-
-  prometheus:
-    image: prom/prometheus:latest
-    volumes:
-      - ./infra/prometheus/prometheus.yml:/etc/prometheus/prometheus.yml
-    restart: unless-stopped
-
-  grafana:
-    image: grafana/grafana:latest
-    ports:
-      - "3000:3000"
-    volumes:
-      - grafana_data:/var/lib/grafana
-    restart: unless-stopped
-
-  loki:
-    image: grafana/loki:latest
-    restart: unless-stopped
-```
-
-Ogni servizio espone `/metrics` (Prometheus) e scrive log strutturati (JSON) raccolti da Loki.
-
----
-
-## 10. Sizing VPS Minimo Consigliato (MVP)
-
-| Componente | CPU | RAM | Note |
-|---|---|---|---|
-| Traefik | 0.25 | 128MB | |
-| Auth Service ×2 | 0.25 ×2 | 128MB ×2 | Stateless, leggero |
-| Chat Service ×2 | 1.0 ×2 | 1GB ×2 | WS + streaming LLM |
-| Agent Service ×2 | 0.75 ×2 | 512MB ×2 | Batch LLM, CPU-bound |
-| Billing Service | 0.25 | 128MB | |
-| PostgreSQL | 1.0 | 1GB | |
-| Redis | 0.25 | 256MB | |
-| Qdrant | 0.5 | 512MB | |
-| **Totale MVP** | **~5.5 vCPU** | **~5 GB** | |
-
-**Raccomandazione**: VPS con **8 vCPU / 16 GB RAM** per avere margine. Hetzner CPX41 (~€30/mese) o equivalente. Senza Storage/Plugin si risparmia ~1 vCPU e 512MB rispetto alla versione completa.
-
----
-
-## Riepilogo Architettura MVP
-
-| Servizio | Repliche | Proprietario di |
-|---|---|---|
-| **Traefik** | 1 | Routing, TLS, sticky sessions |
-| **Auth Service** | 2 | JWT RS256, registrazione, login, profilo |
-| **Chat Service** | 2–N | WebSocket, home/floating chat, streaming |
-| **Agent Service** | 2–N | Batch processing, directory scan, agent setup |
-| **Billing Service** | 1 | Stripe, subscriptions, tier management |
-
-| Decisione | Scelta | Motivazione |
-|---|---|---|
-| API Gateway | Traefik | Nativo Docker, WebSocket support, service discovery automatico |
-| JWT | RS256 (asimmetrico) | Verifica distribuita senza contattare Auth Service |
-| Tier check | Claim nel JWT | Ogni servizio verifica localmente, zero roundtrip |
-| WebSocket scaling | Redis pub/sub + sticky cookies | Cross-instance tool-call routing |
-| Chat ↔ Agent split | Servizi separati | Batch CPU-bound non impatta real-time chat |
-| Agent → Device comms | Redis pub/sub via Chat Service | Agent non possiede la WS, usa un relay |
-| Rate limiting | Redis contatori distribuiti | Sliding window condivisa tra repliche |
-| Database | PostgreSQL condiviso | Semplicità MVP; split DB futuro facile |
-| TLS | Cloudflare Origin Certificate | Zero maintenance |
-| Orchestrazione | Docker Compose | Sufficiente per un singolo VPS |
-| Storage / Plugin | Post-MVP | Non critici per il lancio |

requirements.txt

@@ -32,4 +32,10 @@ google-auth-oauthlib>=1.2.0
 google-auth-httplib2>=0.2.0
 msal>=1.28.0
 cryptography>=42.0.0
+pgvector>=0.2.5
+langfuse>=2.0.0
+beautifulsoup4>=4.12.0
+lxml>=5.0.0
+PyYAML>=6.0.0
+apscheduler>=3.10.0
 ruff>=0.8.0

tests/conftest.py

@@ -6,26 +6,21 @@ a per-test session, and a FastAPI ``TestClient`` wired to use it.
 
 from __future__ import annotations
 
-import json
-import os
 import time
 import uuid
 from collections.abc import AsyncGenerator, Generator
-from unittest.mock import patch
 
-import boto3
 import pytest
 import pytest_asyncio
 from fastapi.testclient import TestClient
 from jose import jwt
-from moto import mock_aws
 from sqlalchemy import StaticPool, event
 from sqlalchemy.ext.asyncio import AsyncSession, async_sessionmaker, create_async_engine
 
 from app.config.settings import settings
 from app.db import Base, get_session
 from app.main import app
-from app.models import Plugin, Subscription, User
+from app.models import Subscription, User
 
 # ── Fixed test user IDs (one per tier) ───────────────────────────────
 
@@ -109,79 +104,6 @@ def client(db_session: AsyncSession) -> Generator[TestClient, None, None]:   # n
     app.dependency_overrides.pop(get_session, None)
 
 
-# ── Seed data helpers ────────────────────────────────────────────────
-
-_SEED_PLUGINS = [
-    Plugin(
-        id="plugin-github-sync",
-        name="GitHub Sync",
-        description="Sync tasks with GitHub Issues and pull requests.",
-        version="1.0.0",
-        author_name="Adiuva",
-        category="productivity",
-        price_cents=0,
-        permissions=json.dumps(["read:tasks", "write:tasks"]),
-        status="approved",
-        s3_package_key="plugins/plugin-github-sync/1.0.0/package.zip",
-        install_count=0,
-        avg_rating=0.0,
-    ),
-    Plugin(
-        id="plugin-slack-notify",
-        name="Slack Notifier",
-        description="Post task and timeline updates to Slack channels.",
-        version="1.2.0",
-        author_name="Adiuva",
-        category="communication",
-        price_cents=499,
-        permissions=json.dumps(["read:tasks", "read:timelines"]),
-        status="approved",
-        s3_package_key="plugins/plugin-slack-notify/1.2.0/package.zip",
-        install_count=0,
-        avg_rating=0.0,
-    ),
-    Plugin(
-        id="plugin-time-tracker",
-        name="Time Tracker",
-        description="Track time spent on tasks with automatic reporting.",
-        version="0.9.1",
-        author_name="Third Party",
-        category="productivity",
-        price_cents=999,
-        permissions=json.dumps(["read:tasks", "write:tasks"]),
-        status="approved",
-        s3_package_key="plugins/plugin-time-tracker/0.9.1/package.zip",
-        install_count=0,
-        avg_rating=0.0,
-    ),
-]
-
-
-@pytest_asyncio.fixture
-async def seed_plugins(db_session: AsyncSession) -> list[Plugin]:
-    """Insert the 3 default approved plugins and return them."""
-    plugins = []
-    for template in _SEED_PLUGINS:
-        p = Plugin(
-            id=template.id,
-            name=template.name,
-            description=template.description,
-            version=template.version,
-            author_name=template.author_name,
-            category=template.category,
-            price_cents=template.price_cents,
-            permissions=template.permissions,
-            status=template.status,
-            s3_package_key=template.s3_package_key,
-            install_count=template.install_count,
-            avg_rating=template.avg_rating,
-        )
-        db_session.add(p)
-        plugins.append(p)
-    await db_session.commit()
-    return plugins
-
-
 # ── JWT helpers ──────────────────────────────────────────────────────
 
 
@@ -212,24 +134,21 @@ def auth_header(tier: str = "power", user_id: str | None = None) -> dict[str, st
     return {"Authorization": f"Bearer {make_jwt(tier, user_id)}"}
 
 
-# ── S3 mock fixture ──────────────────────────────────────────────────
+# ── CLI options ───────────────────────────────────────────────────────
 
-S3_TEST_BUCKET = "test-bucket"
+def pytest_addoption(parser):
-S3_TEST_REGION = "us-east-1"
+    parser.addoption(
-
+        "--preprocess-dir",
-
+        default=None,
-@pytest.fixture
+        help="Override fixture folder for preprocessor tests (must contain cases.yaml + data/)",
-def s3_bucket():
+    )
-    """Create a mocked S3 bucket via moto and patch BlobStore settings."""
+    parser.addoption(
-    with mock_aws():
+        "--runner-dir",
-        os.environ.setdefault("AWS_ACCESS_KEY_ID", "testing")
+        default=None,
-        os.environ.setdefault("AWS_SECRET_ACCESS_KEY", "testing")
+        help="Override fixture folder for agent_runner_v2 eval tests (must contain cases.yaml + data/)",
-        os.environ.setdefault("AWS_DEFAULT_REGION", S3_TEST_REGION)
+    )
-        client = boto3.client("s3", region_name=S3_TEST_REGION)
+    parser.addoption(
-        client.create_bucket(Bucket=S3_TEST_BUCKET)
+        "--journey-dir",
-        with patch("app.storage.blob_store.settings") as mock_settings:
+        default=None,
-            mock_settings.S3_BUCKET = S3_TEST_BUCKET
+        help="Override fixture folder for journey_v2 eval tests (must contain cases.yaml + data/)",
-            mock_settings.S3_REGION = S3_TEST_REGION
+    )
-            mock_settings.AWS_ACCESS_KEY_ID = "testing"
-            mock_settings.AWS_SECRET_ACCESS_KEY = "testing"
-            yield S3_TEST_BUCKET

tests/fixtures/agent_runner_v2/cases.yaml

@@ -0,0 +1,86 @@
+# Agent Runner V2 — eval test cases (Step 2, requires real LLM)
+#
+# Each case drives one parametrized `test_eval_runner` invocation.
+#
+# Keys
+# ----
+# id: str                     unique identifier shown in pytest output
+# description: str            human-readable label
+# file: str                   filename inside data/
+# file_path: str              path reported to the executor (affects project-matching via filename)
+# projects: [alpha|beta]      symbolic project names resolved by the test helper
+#
+# Optional pre-existing records (dedup tests)
+# existing_tasks:             list of {id, title, status, priority}
+# existing_notes:             list of {id, title, content}
+# existing_timelines:         list of {id, title, date}
+#
+# Assertions (one or more)
+# expect_insert: <table>      at least 1 insert row in this table (tasks|notes|timelines)
+# expect_no_insert: true      zero inserts in any table
+# expect_project_id: <id>     any insert must carry this projectId
+# expect_dedup: true          task inserts == 0 OR task updates >= 1 (dedup check)
+#
+# Langfuse
+# score_name: str             observation score name
+
+- id: "2.1"
+  description: "Action email → create_task"
+  file: email_action.html
+  file_path: /emails/ProjectAlpha_action.html
+  projects: [alpha, beta]
+  expect_insert: tasks
+  score_name: runner.email_to_task
+
+- id: "2.2"
+  description: "Informational email → create_note"
+  file: email_info.html
+  file_path: /emails/ProjectAlpha_info.html
+  projects: [alpha, beta]
+  expect_insert: notes
+  score_name: runner.email_to_note
+
+- id: "2.3"
+  description: "Email with meeting date → create_timeline"
+  file: email_date.html
+  file_path: /emails/ProjectAlpha_kickoff.html
+  projects: [alpha, beta]
+  expect_insert: timelines
+  score_name: runner.email_to_timeline
+
+- id: "2.4"
+  description: "Filename contains project name → correct project assigned"
+  file: email_action.html
+  file_path: /emails/ProjectAlpha_report.html
+  projects: [alpha, beta]
+  expect_project_id: proj-alpha
+  score_name: runner.project_filename
+
+- id: "2.5"
+  description: "Email body mentions project → correct project assigned"
+  file: email_action.html
+  file_path: /emails/email_001.html
+  projects: [alpha, beta]
+  expect_project_id: proj-alpha
+  score_name: runner.project_content
+
+- id: "2.6"
+  description: "Newsletter + global rule no-project → no creates"
+  file: email_no_project.html
+  file_path: /emails/newsletter.html
+  projects: [alpha, beta]
+  expect_no_insert: true
+  score_name: runner.no_project
+
+- id: "2.7"
+  description: "Existing task with same title → dedup (update not create)"
+  file: email_action.html
+  file_path: /emails/ProjectAlpha_followup.html
+  projects: [alpha]
+  existing_tasks:
+    - id: task-existing
+      title: Fix the login bug
+      status: todo
+      priority: medium
+  expect_dedup: true
+  score_name: runner.dedup

tests/fixtures/agent_runner_v2/data/email_action.html

@@ -0,0 +1,7 @@
+<html><head></head><body>
+<p><b>From:</b> boss@company.com</p>
+<p><b>To:</b> dev@company.com</p>
+<p><b>Subject:</b> Fix the login bug</p>
+<p><b>Date:</b> 2026-04-07</p>
+<p>Hi,<br>Please fix the login bug in Project Alpha by Friday. High priority!</p>
+</body></html>

tests/fixtures/agent_runner_v2/data/email_date.html

@@ -0,0 +1,5 @@
+<html><head></head><body>
+<p><b>From:</b> pm@company.com</p>
+<p><b>Subject:</b> Project Alpha kick-off meeting</p>
+<p>The kick-off meeting for Project Alpha is scheduled for 2026-04-15 at 10:00.</p>
+</body></html>

tests/fixtures/agent_runner_v2/data/email_info.html

@@ -0,0 +1,7 @@
+<html><head></head><body>
+<p><b>From:</b> pm@company.com</p>
+<p><b>To:</b> team@company.com</p>
+<p><b>Subject:</b> FYI: New policy for Project Alpha</p>
+<p>Just a heads-up that starting next week all code reviews must be done
+within 24 hours for Project Alpha. No action needed from you now.</p>
+</body></html>

tests/fixtures/agent_runner_v2/data/email_no_project.html

@@ -0,0 +1,5 @@
+<html><head></head><body>
+<p><b>From:</b> newsletter@ads.com</p>
+<p><b>Subject:</b> Weekly newsletter</p>
+<p>Check out our latest deals on electronics!</p>
+</body></html>

tests/fixtures/journey_v2/cases.yaml

@@ -0,0 +1,19 @@
+# Journey V2 eval test cases — Step 4
+#
+# Only case 4.1 is kept as an automated eval. Cases 4.2–4.5 (multi-turn
+# conversations that expect the LLM to produce a complete AgentConfig)
+# are non-deterministic and tested manually — results tracked in Langfuse.
+#
+# Assertion keys:
+#   expect_question: true → first reply must contain "?"
+
+- id: "4.1"
+  description: "Journey start explores directory, first reply contains a question"
+  directory: "/test/emails"
+  data_types: ["tasks", "notes", "timelines"]
+  directory_files:
+    - path: "/test/emails/outlook_export_2024.html"
+      content_file: "email_action.html"
+  user_messages: []
+  score_name: "journey.start"
+  expect_question: true

tests/fixtures/journey_v2/data/email_action.html

@@ -0,0 +1,23 @@
+<!DOCTYPE html>
+<html>
+<head>
+  <meta charset="UTF-8">
+  <title>Email: Fix the login bug</title>
+  <style>body { font-family: Arial; } .header { color: #666; }</style>
+</head>
+<body>
+  <div class="header">
+    <p><strong>From:</strong> boss@company.com</p>
+    <p><strong>To:</strong> dev@company.com</p>
+    <p><strong>Subject:</strong> Fix the login bug</p>
+    <p><strong>Date:</strong> Mon, 7 Apr 2026 09:15:00 +0000</p>
+  </div>
+  <div class="body">
+    <p>Hi,</p>
+    <p>Please fix the login bug in Project Alpha as soon as possible.
+    Users are reporting that they can't log in with their Google accounts.
+    This is blocking the whole team. Please resolve it by Friday.</p>
+    <p>Thanks,<br>Boss</p>
+  </div>
+</body>
+</html>

tests/fixtures/journey_v2/data/email_info.html

@@ -0,0 +1,23 @@
+<!DOCTYPE html>
+<html>
+<head>
+  <meta charset="UTF-8">
+  <title>Email: New policy update</title>
+  <style>body { font-family: Arial; }</style>
+</head>
+<body>
+  <div class="header">
+    <p><strong>From:</strong> hr@company.com</p>
+    <p><strong>To:</strong> all@company.com</p>
+    <p><strong>Subject:</strong> FYI: New remote work policy effective May 1</p>
+    <p><strong>Date:</strong> Tue, 8 Apr 2026 10:00:00 +0000</p>
+  </div>
+  <div class="body">
+    <p>Hi everyone,</p>
+    <p>Just a heads-up that starting May 1, 2026 the company will be moving to
+    a hybrid work model. You will be expected to come into the office at least
+    two days per week. More details will follow in the employee handbook.</p>
+    <p>Best,<br>HR Team</p>
+  </div>
+</body>
+</html>

tests/fixtures/preprocessors/cases.yaml

@@ -0,0 +1,68 @@
+# Preprocessor test cases
+#
+# detect: <expected_type>   → chiama detect_content_type(filename, content)
+# process: <content_type>   → chiama preprocess(content_type, content)
+#
+# Sorgente: file: <nome in data/>  oppure  generate: binary_noise
+#
+# Assertions piatte (solo per process):
+#   no_html: true           clean_text senza tag HTML
+#   min_chars: N            len(clean_text) >= N
+#   ratio_lt: F             len(clean) / len(raw) < F
+#   has_meta: [k, ...]      chiavi presenti in metadata
+#   contains: str | [str]   substring(s) presenti in clean_text
+#   excludes: str | [str]   substring(s) assenti da clean_text
+#   content_type: str       result.content_type == questo valore
+
+- id: "1.1"
+  file: email_action.html
+  detect: email_html
+
+- id: "1.2"
+  file: generic_page.html
+  detect: generic_html
+
+- id: "1.3"
+  file: notes.txt
+  detect: plain_text
+
+- id: "1.4"
+  file: archive.xyz
+  generate: binary_noise
+  detect: unknown
+
+- id: "1.5"
+  file: email_action.html
+  process: email_html
+  no_html: true
+  min_chars: 50
+  ratio_lt: 0.8
+
+- id: "1.6"
+  file: email_action.html
+  process: email_html
+  has_meta: [subject, from]
+
+- id: "1.7"
+  file: email_thread.html
+  process: email_html
+  contains: "Sure, I'll handle the deploy"
+  excludes: "Let's plan the deploy"
+
+- id: "1.8"
+  file: email_single.html
+  process: email_html
+  contains: "deploy is done"
+
+- id: "1.9"
+  file: email_heavy.html
+  process: email_html
+  no_html: true
+  min_chars: 30
+  excludes: [border-collapse, font-size]
+
+- id: "1.10"
+  file: fallback.txt
+  process: unknown
+  min_chars: 1
+  content_type: unknown

tests/fixtures/preprocessors/data/email_action.html

@@ -0,0 +1,25 @@
+<!DOCTYPE html>
+<html>
+<head>
+  <title>Fix the login bug</title>
+  <style>
+    body { font-family: Arial, sans-serif; color: #333; margin: 0; padding: 20px; }
+    .header { background: #f5f5f5; padding: 10px; border-bottom: 1px solid #ddd; }
+    .body { padding: 20px; }
+  </style>
+</head>
+<body>
+  <div class="header">
+    <p><strong>From:</strong> boss@company.com</p>
+    <p><strong>To:</strong> dev@company.com</p>
+    <p><strong>Subject:</strong> Fix the login bug</p>
+    <p><strong>Date:</strong> Mon, 7 Apr 2026 09:00:00 +0200</p>
+  </div>
+  <div class="body">
+    <p>Hi,</p>
+    <p>Please fix the login bug by Friday. It is blocking the release.</p>
+    <p>Priority: high. Let me know if you need anything.</p>
+    <p>Thanks,<br>Boss</p>
+  </div>
+</body>
+</html>

tests/fixtures/preprocessors/data/email_heavy.html

@@ -0,0 +1,49 @@
+<!DOCTYPE html>
+<html>
+<head>
+<style>
+  table { border-collapse: collapse; width: 100%; max-width: 600px; margin: 0 auto; }
+  td { padding: 8px 12px; border: 1px solid #dddddd; font-size: 12px; color: #444444; }
+  .header-row { background-color: #003366; color: #ffffff; font-weight: bold; }
+  .label-col { background-color: #f0f0f0; width: 80px; font-weight: bold; }
+  .footer-row { font-size: 10px; color: #999999; text-align: center; }
+</style>
+</head>
+<body bgcolor="#eeeeee">
+<center>
+<table cellpadding="0" cellspacing="0">
+  <tr class="header-row">
+    <td colspan="2">Company Internal Update</td>
+  </tr>
+  <tr>
+    <td class="label-col">From:</td>
+    <td>newsletter@corp.com</td>
+  </tr>
+  <tr>
+    <td class="label-col">Subject:</td>
+    <td>Q1 Results Update</td>
+  </tr>
+  <tr>
+    <td class="label-col">Date:</td>
+    <td>Apr 7, 2026</td>
+  </tr>
+  <tr>
+    <td colspan="2">
+      <table width="100%" cellpadding="10">
+        <tr>
+          <td>
+            <p style="font-size:14px; font-weight:bold;">Dear Team,</p>
+            <p>Q1 results are in. Revenue up 15% year-over-year.</p>
+            <p>Please review the attached report and share any feedback by EOW.</p>
+          </td>
+        </tr>
+      </table>
+    </td>
+  </tr>
+  <tr class="footer-row">
+    <td colspan="2">Confidential — do not forward outside the company.</td>
+  </tr>
+</table>
+</center>
+</body>
+</html>

tests/fixtures/preprocessors/data/email_single.html

@@ -0,0 +1,8 @@
+<!DOCTYPE html>
+<html><body>
+  <p><strong>From:</strong> alice@co.com</p>
+  <p><strong>To:</strong> team@co.com</p>
+  <p><strong>Subject:</strong> Quick update</p>
+  <p><strong>Date:</strong> Tue, 7 Apr 2026 10:30:00 +0200</p>
+  <p>The deploy is done. Everything looks good. No issues so far.</p>
+</body></html>

tests/fixtures/preprocessors/data/email_thread.html

@@ -0,0 +1,24 @@
+<!DOCTYPE html>
+<html><body>
+  <div class="message-latest">
+    <p><strong>From:</strong> alice@co.com</p>
+    <p><strong>Subject:</strong> Re: Re: Deploy plan</p>
+    <p>Sure, I'll handle the deploy.</p>
+  </div>
+
+  <p>On Mon, Apr 6, 2026 at 3:00 PM, Bob &lt;bob@co.com&gt; wrote:</p>
+  <blockquote>
+    <p>From: bob@co.com</p>
+    <p>Can you handle the deploy?</p>
+    <p>On Sun, Apr 5, 2026 at 1:00 PM, Alice &lt;alice@co.com&gt; wrote:</p>
+    <blockquote>
+      <p>From: alice@co.com</p>
+      <p>Let's plan the deploy for Monday.</p>
+      <p>On Sat, Apr 4, 2026 at 11:00 AM, Charlie &lt;charlie@co.com&gt; wrote:</p>
+      <blockquote>
+        <p>From: charlie@co.com</p>
+        <p>We need to schedule the deploy. What day works?</p>
+      </blockquote>
+    </blockquote>
+  </blockquote>
+</body></html>

tests/fixtures/preprocessors/data/fallback.txt

@@ -0,0 +1,3 @@
+random text content without any structure
+line two with some words
+line three and more content here

tests/fixtures/preprocessors/data/generic_page.html

@@ -0,0 +1,35 @@
+<!DOCTYPE html>
+<html lang="en">
+<head>
+  <meta charset="UTF-8">
+  <title>My Web App</title>
+  <link rel="stylesheet" href="styles.css">
+</head>
+<body>
+  <nav>
+    <a href="/">Home</a>
+    <a href="/about">About</a>
+    <a href="/contact">Contact</a>
+  </nav>
+  <main>
+    <header>
+      <h1>Welcome to My App</h1>
+    </header>
+    <article>
+      <p>This is a generic web page with no email headers.</p>
+      <p>It has navigation, main content, and a footer.</p>
+    </article>
+    <section>
+      <h2>Features</h2>
+      <ul>
+        <li>Fast</li>
+        <li>Reliable</li>
+        <li>Secure</li>
+      </ul>
+    </section>
+  </main>
+  <footer>
+    <p>&copy; 2026 My App</p>
+  </footer>
+</body>
+</html>

tests/fixtures/preprocessors/data/notes.txt

@@ -0,0 +1,15 @@
+Meeting notes - April 7, 2026
+
+Attendees: Alice, Bob, Charlie
+
+Discussion points:
+- Deploy scheduled for Friday
+- Bug fix for login must be completed by Thursday
+- Review Q1 numbers before EOW
+
+Action items:
+- Alice: fix login bug
+- Bob: prepare deploy checklist
+- Charlie: send Q1 report
+
+Next meeting: April 14, 2026

tests/test_agent_runner.py

@@ -1,810 +0,0 @@
-"""Tests for Step 3.4: agent_runner module.
-
-Coverage:
-  Unit:
-    - _is_overdue      — cron schedule overdue detection
-    - _extract_items_from_content — LLM extraction + JSON parsing + validation
-    - _send_insert_to_client      — tool_call frame construction + timeout
-    - run_local_agent             — end-to-end local agent happy path
-    - run_local_agent             — device offline path
-    - run_local_agent             — file-read timeout path
-    - run_local_agent             — LLM extraction error path
-    - run_cloud_agent             — stub returns error immediately
-    - trigger_pending_runs        — skipped when config is client-owned
-    - trigger_pending_runs        — non-overdue skipped
-    - trigger_pending_runs        — device_id filter for local agents
-
-    Integration:
-        - POST /agents/can-create     — billing eligibility check
-        - POST /agents/trigger        — creates run log + dispatches background task
-"""
-
-from __future__ import annotations
-
-import asyncio
-import json
-import uuid
-from datetime import datetime, timezone
-from unittest.mock import AsyncMock, MagicMock, patch
-
-import pytest
-import pytest_asyncio
-
-from app.core.agent_runner import (
-    _extract_items_from_content,
-    _is_overdue,
-    _send_insert_to_client,
-    run_cloud_agent,
-    run_local_agent,
-    trigger_pending_runs,
-)
-from app.core.device_manager import DeviceConnectionManager
-from app.db import get_session
-from app.main import app
-from app.models import AgentRunLog, CloudAgentConfig, LocalAgentConfig
-from tests.conftest import TEST_USER_IDS, auth_header
-
-# ---------------------------------------------------------------------------
-# Helpers
-# ---------------------------------------------------------------------------
-
-_FREE_UID = TEST_USER_IDS["free"]
-_PRO_UID = TEST_USER_IDS["pro"]
-
-
-def _make_local_config(user_id: str = _FREE_UID, device_id: str = "dev-001") -> LocalAgentConfig:
-    return LocalAgentConfig(
-        id=str(uuid.uuid4()),
-        user_id=user_id,
-        device_id=device_id,
-        name="Test Local Agent",
-        directory_paths=["/home/user/emails"],
-        data_types=["tasks", "notes"],
-        prompt_template="Extract tasks and notes from this document.",
-        file_extensions=[".txt", ".eml"],
-        schedule_cron="0 */6 * * *",
-        enabled=True,
-        last_run_at=None,
-    )
-
-
-def _make_cloud_config(user_id: str = _FREE_UID) -> CloudAgentConfig:
-    return CloudAgentConfig(
-        id=str(uuid.uuid4()),
-        user_id=user_id,
-        provider="gmail",
-        name="Test Gmail Agent",
-        data_types=["tasks"],
-        prompt_template="Extract tasks from email.",
-        schedule_cron="0 */6 * * *",
-        enabled=True,
-        last_run_at=None,
-    )
-
-
-def _make_run_log(agent_id: str, agent_type: str = "local", user_id: str = _FREE_UID) -> AgentRunLog:
-    return AgentRunLog(
-        id=str(uuid.uuid4()),
-        agent_id=agent_id,
-        agent_type=agent_type,
-        user_id=user_id,
-        status="running",
-        started_at=datetime.now(timezone.utc),
-    )
-
-
-def _make_manager(user_id: str = _FREE_UID, device_id: str = "dev-001") -> DeviceConnectionManager:
-    mgr = DeviceConnectionManager()
-    ws = MagicMock()
-    ws.send_text = AsyncMock()
-    mgr.register(user_id, device_id, ws)
-    return mgr
-
-
-# ---------------------------------------------------------------------------
-# _is_overdue
-# ---------------------------------------------------------------------------
-
-def test_is_overdue_never_run():
-    """An agent that has never run is always overdue."""
-    assert _is_overdue("0 */6 * * *", None) is True
-
-
-def test_is_overdue_very_recently_run():
-    """An agent that just ran is not overdue."""
-    last = datetime.now(timezone.utc)
-    assert _is_overdue("0 */6 * * *", last) is False
-
-
-def test_is_overdue_long_ago():
-    """An agent last run 2 days ago with a 6-hour schedule is overdue."""
-    from datetime import timedelta
-    last = datetime.now(timezone.utc) - timedelta(days=2)
-    assert _is_overdue("0 */6 * * *", last) is True
-
-
-def test_is_overdue_invalid_cron_returns_false():
-    """Unparseable cron must not raise and should return False (fail-safe)."""
-    assert _is_overdue("not a cron", None) is False
-
-
-def test_is_overdue_naive_datetime():
-    """Naive datetime objects are handled without raising."""
-    from datetime import timedelta
-    last = datetime.utcnow() - timedelta(days=1)  # naive
-    # Should not raise.
-    result = _is_overdue("0 */6 * * *", last)
-    assert isinstance(result, bool)
-
-
-# ---------------------------------------------------------------------------
-# _extract_items_from_content
-# ---------------------------------------------------------------------------
-
-
-@pytest.mark.asyncio
-async def test_extract_items_happy_path():
-    """LLM returns valid JSON array; items with allowed tables are returned."""
-    mock_llm = MagicMock()
-    mock_response = MagicMock()
-    mock_response.content = json.dumps([
-        {"table": "tasks", "data": {"title": "Buy milk", "priority": "high"}},
-        {"table": "notes", "data": {"title": "Meeting recap", "content": "Discussed roadmap"}},
-    ])
-    mock_llm.ainvoke = AsyncMock(return_value=mock_response)
-
-    with patch("app.core.agent_runner.get_llm", return_value=mock_llm):
-        items = await _extract_items_from_content(
-            "Extract tasks and notes.",
-            "Email body: Buy milk urgently. Notes from meeting: discussed roadmap.",
-            ["tasks", "notes"],
-        )
-
-    assert len(items) == 2
-    assert items[0]["table"] == "tasks"
-    assert items[0]["data"]["title"] == "Buy milk"
-    assert items[1]["table"] == "notes"
-
-
-@pytest.mark.asyncio
-async def test_extract_items_strips_forbidden_fields():
-    """Fields like id, createdAt, isAiSuggested must be stripped from extracted data."""
-    mock_llm = MagicMock()
-    mock_response = MagicMock()
-    mock_response.content = json.dumps([
-        {
-            "table": "tasks",
-            "data": {
-                "title": "Review PR",
-                "id": "should-be-removed",
-                "createdAt": 99999,
-                "isAiSuggested": 0,
-                "isApproved": 1,
-            },
-        }
-    ])
-    mock_llm.ainvoke = AsyncMock(return_value=mock_response)
-
-    with patch("app.core.agent_runner.get_llm", return_value=mock_llm):
-        items = await _extract_items_from_content("Extract tasks.", "Review the PR.", ["tasks"])
-
-    assert len(items) == 1
-    data = items[0]["data"]
-    assert "id" not in data
-    assert "createdAt" not in data
-    assert "isAiSuggested" not in data
-    assert "isApproved" not in data
-    assert data["title"] == "Review PR"
-
-
-@pytest.mark.asyncio
-async def test_extract_items_invalid_json_returns_empty():
-    """LLM returning invalid JSON must return empty list without raising."""
-    mock_llm = MagicMock()
-    mock_response = MagicMock()
-    mock_response.content = "Sorry, I cannot extract anything."
-    mock_llm.ainvoke = AsyncMock(return_value=mock_response)
-
-    with patch("app.core.agent_runner.get_llm", return_value=mock_llm):
-        items = await _extract_items_from_content("Extract tasks.", "content", ["tasks"])
-
-    assert items == []
-
-
-@pytest.mark.asyncio
-async def test_extract_items_disallowed_table_filtered():
-    """Items whose table is not in data_types are discarded."""
-    mock_llm = MagicMock()
-    mock_response = MagicMock()
-    mock_response.content = json.dumps([
-        {"table": "tasks", "data": {"title": "Valid task"}},
-        {"table": "projects", "data": {"name": "Should be filtered"}},
-    ])
-    mock_llm.ainvoke = AsyncMock(return_value=mock_response)
-
-    with patch("app.core.agent_runner.get_llm", return_value=mock_llm):
-        # Only "tasks" is in data_types — "projects" should be filtered.
-        items = await _extract_items_from_content("Extract.", "content", ["tasks"])
-
-    assert len(items) == 1
-    assert items[0]["table"] == "tasks"
-
-
-@pytest.mark.asyncio
-async def test_extract_items_empty_data_types_returns_empty():
-    """If no allowed data_types match, skip LLM call and return immediately."""
-    mock_llm = MagicMock()
-    mock_llm.ainvoke = AsyncMock()
-
-    with patch("app.core.agent_runner.get_llm", return_value=mock_llm):
-        items = await _extract_items_from_content("Extract.", "content", [])
-
-    mock_llm.ainvoke.assert_not_called()
-    assert items == []
-
-
-@pytest.mark.asyncio
-async def test_extract_items_llm_error_propagates():
-    """LLM API errors propagate so the caller (run_local_agent) can record them."""
-    mock_llm = MagicMock()
-    mock_llm.ainvoke = AsyncMock(side_effect=RuntimeError("API unavailable"))
-
-    with patch("app.core.agent_runner.get_llm", return_value=mock_llm):
-        with pytest.raises(RuntimeError, match="API unavailable"):
-            await _extract_items_from_content("Extract tasks.", "content", ["tasks"])
-
-
-# ---------------------------------------------------------------------------
-# _send_insert_to_client
-# ---------------------------------------------------------------------------
-
-
-@pytest.mark.asyncio
-async def test_send_insert_to_client_happy_path():
-    """Frame is sent with isAiSuggested/isApproved added; result is returned."""
-    mgr = _make_manager()
-
-    sent_payloads: list[dict] = []
-    original_send = mgr.send_frame
-
-    async def _capture_send(uid: str, frame: dict) -> None:
-        sent_payloads.append(frame)
-        # Immediately resolve the pending call with a success result.
-        call_id = frame["id"]
-        mgr.resolve_pending_call(uid, call_id, {"row": {"id": "new-id", "title": "Buy milk"}})
-
-    mgr.send_frame = _capture_send  # type: ignore[method-assign]
-
-    result = await _send_insert_to_client(
-        _FREE_UID, "tasks", {"title": "Buy milk", "priority": "high"}, mgr
-    )
-
-    assert len(sent_payloads) == 1
-    payload = sent_payloads[0]
-    assert payload["action"] == "insert"
-    assert payload["table"] == "tasks"
-    assert payload["data"]["title"] == "Buy milk"
-    assert payload["data"]["isAiSuggested"] == 1
-    assert payload["data"]["isApproved"] == 0
-    assert result["row"]["title"] == "Buy milk"
-
-
-@pytest.mark.asyncio
-async def test_send_insert_to_client_timeout():
-    """asyncio.TimeoutError is raised when Electron does not respond."""
-    mgr = _make_manager()
-
-    async def _slow_send(uid: str, frame: dict) -> None:
-        # Never resolve the pending call.
-        pass
-
-    mgr.send_frame = _slow_send  # type: ignore[method-assign]
-
-    with patch("app.core.agent_runner._INSERT_TIMEOUT", 0.05):
-        with pytest.raises(asyncio.TimeoutError):
-            await _send_insert_to_client(_FREE_UID, "tasks", {"title": "X"}, mgr)
-
-
-# ---------------------------------------------------------------------------
-# run_local_agent
-# ---------------------------------------------------------------------------
-
-
-@pytest.mark.asyncio
-async def test_run_local_agent_device_offline():
-    """run_local_agent marks run as error when device is offline."""
-    config = _make_local_config()
-    run_log = _make_run_log(config.id)
-    mgr = DeviceConnectionManager()  # Empty — no device registered.
-
-    with patch("app.core.agent_runner._finalize_run", new_callable=AsyncMock) as mock_finalize:
-        await run_local_agent(_FREE_UID, config, run_log, mgr)
-
-    mock_finalize.assert_called_once()
-    _args, kwargs = mock_finalize.call_args
-    assert kwargs["status"] == "error"
-    assert any("not connected" in e for e in kwargs["errors"])
-
-
-@pytest.mark.asyncio
-async def test_run_local_agent_happy_path():
-    """End-to-end: files received, LLM extracts one task, insert sent + ack'd."""
-    config = _make_local_config()
-    run_log = _make_run_log(config.id)
-    mgr = _make_manager()
-
-    # Build a fake agent_data frame (will be queued after send).
-    file_frame = {
-        "type": "agent_data",
-        "run_id": run_log.id,
-        "files": [{"path": "/email.eml", "content": "Urgent: fix the bug by Friday."}],
-    }
-    agent_complete_frame = None  # sentinel
-
-    sent_frames: list[dict] = []
-
-    async def _mock_send(uid: str, frame: dict) -> None:
-        sent_frames.append(frame)
-        if frame.get("type") == "agent_run":
-            # Simulate Electron responding with file data then agent_complete.
-            q = mgr.get_agent_data_queue(uid, frame["run_id"])
-            await q.put(file_frame)
-            await q.put(agent_complete_frame)
-        elif frame.get("type") == "tool_call":
-            # Resolve the pending insert immediately.
-            mgr.resolve_pending_call(uid, frame["id"], {"row": {"id": "new-task", "title": "Fix the bug"}})
-
-    mgr.send_frame = _mock_send  # type: ignore[method-assign]
-
-    mock_llm = MagicMock()
-    mock_response = MagicMock()
-    mock_response.content = json.dumps([
-        {"table": "tasks", "data": {"title": "Fix the bug", "priority": "high"}}
-    ])
-    mock_llm.ainvoke = AsyncMock(return_value=mock_response)
-
-    with patch("app.core.agent_runner.get_llm", return_value=mock_llm), \
-         patch("app.core.agent_runner._finalize_run", new_callable=AsyncMock) as mock_finalize:
-        await run_local_agent(_FREE_UID, config, run_log, mgr)
-
-    mock_finalize.assert_called_once()
-    _args, kwargs = mock_finalize.call_args
-    assert kwargs["status"] == "success"
-    assert kwargs["items_processed"] == 1
-    assert kwargs["items_created"] == 1
-    assert kwargs["errors"] == []
-    assert kwargs["update_config_last_run"] is False
-
-    # Verify agent_run frame was sent.
-    agent_run_frames = [f for f in sent_frames if f.get("type") == "agent_run"]
-    assert len(agent_run_frames) == 1
-    assert agent_run_frames[0]["agent_id"] == config.id
-    assert "paths" in agent_run_frames[0]["config"]
-
-    # Verify insert frame was sent with AI flags.
-    insert_frames = [f for f in sent_frames if f.get("type") == "tool_call"]
-    assert len(insert_frames) == 1
-    assert insert_frames[0]["data"]["isAiSuggested"] == 1
-    assert insert_frames[0]["data"]["isApproved"] == 0
-
-
-@pytest.mark.asyncio
-async def test_run_local_agent_file_read_timeout():
-    """run_local_agent marks run as partial/error when device stops sending files."""
-    config = _make_local_config()
-    run_log = _make_run_log(config.id)
-    mgr = _make_manager()
-
-    async def _mock_send(uid: str, frame: dict) -> None:
-        # Don't put anything in the queue — simulate stalled device.
-        pass
-
-    mgr.send_frame = _mock_send  # type: ignore[method-assign]
-
-    with patch("app.core.agent_runner._FILE_READ_TIMEOUT", 0.1), \
-         patch("app.core.agent_runner._finalize_run", new_callable=AsyncMock) as mock_finalize:
-        await run_local_agent(_FREE_UID, config, run_log, mgr)
-
-    mock_finalize.assert_called_once()
-    _args, kwargs = mock_finalize.call_args
-    assert kwargs["status"] == "error"  # No items created, so error (not partial).
-    assert any("timed out" in e.lower() for e in kwargs["errors"])
-
-
-@pytest.mark.asyncio
-async def test_run_local_agent_llm_extraction_error():
-    """LLM errors per-file are recorded; run continues for remaining files."""
-    config = _make_local_config()
-    run_log = _make_run_log(config.id)
-    mgr = _make_manager()
-
-    file_frame = {
-        "type": "agent_data",
-        "run_id": run_log.id,
-        "files": [
-            {"path": "/file1.eml", "content": "Email one."},
-            {"path": "/file2.eml", "content": "Email two."},
-        ],
-    }
-
-    async def _mock_send(uid: str, frame: dict) -> None:
-        if frame.get("type") == "agent_run":
-            q = mgr.get_agent_data_queue(uid, frame["run_id"])
-            await q.put(file_frame)
-            await q.put(None)  # agent_complete sentinel
-
-    mgr.send_frame = _mock_send  # type: ignore[method-assign]
-
-    mock_llm = MagicMock()
-    mock_llm.ainvoke = AsyncMock(side_effect=RuntimeError("LLM boom"))
-
-    with patch("app.core.agent_runner.get_llm", return_value=mock_llm), \
-         patch("app.core.agent_runner._finalize_run", new_callable=AsyncMock) as mock_finalize:
-        await run_local_agent(_FREE_UID, config, run_log, mgr)
-
-    _args, kwargs = mock_finalize.call_args
-    assert kwargs["status"] == "error"
-    assert kwargs["items_processed"] == 2  # Both files attempted.
-    assert kwargs["items_created"] == 0
-    assert len(kwargs["errors"]) == 2  # One error per file.
-
-
-# ---------------------------------------------------------------------------
-# run_cloud_agent (stub)
-# ---------------------------------------------------------------------------
-
-
-@pytest.mark.asyncio
-async def test_run_cloud_agent_device_offline():
-    """Cloud agent aborts immediately when no device is connected."""
-    config = _make_cloud_config()
-    run_log = _make_run_log(config.id, agent_type="cloud")
-    mgr = DeviceConnectionManager()  # empty — no devices registered
-
-    with patch("app.core.agent_runner._finalize_run", new_callable=AsyncMock) as mock_finalize:
-        await run_cloud_agent(_FREE_UID, config, run_log, mgr)
-
-    mock_finalize.assert_called_once()
-    _, kwargs = mock_finalize.call_args
-    assert kwargs["status"] == "error"
-    assert any("device" in e.lower() or "connected" in e.lower() for e in kwargs["errors"])
-
-
-@pytest.mark.asyncio
-async def test_run_cloud_agent_no_oauth_token():
-    """Cloud agent errors when no OAuth token is stored."""
-    config = _make_cloud_config()
-    config.oauth_token_encrypted = None
-    run_log = _make_run_log(config.id, agent_type="cloud")
-    mgr = _make_manager()
-
-    with patch("app.core.agent_runner._finalize_run", new_callable=AsyncMock) as mock_finalize:
-        await run_cloud_agent(_FREE_UID, config, run_log, mgr)
-
-    _, kwargs = mock_finalize.call_args
-    assert kwargs["status"] == "error"
-    assert any("oauth" in e.lower() or "token" in e.lower() for e in kwargs["errors"])
-
-
-@pytest.mark.asyncio
-async def test_run_cloud_agent_token_decrypt_failure():
-    """Cloud agent errors gracefully when the stored token cannot be decrypted."""
-    config = _make_cloud_config()
-    config.oauth_token_encrypted = "this-is-not-valid-fernet-ciphertext"
-    run_log = _make_run_log(config.id, agent_type="cloud")
-    mgr = _make_manager()
-
-    from cryptography.fernet import Fernet as _Fernet
-    valid_key = _Fernet.generate_key().decode()
-
-    with patch("app.core.agent_runner._finalize_run", new_callable=AsyncMock) as mock_finalize, \
-         patch("app.integrations.settings") as mock_settings:
-        mock_settings.OAUTH_ENCRYPTION_KEY = valid_key
-        await run_cloud_agent(_FREE_UID, config, run_log, mgr)
-
-    _, kwargs = mock_finalize.call_args
-    assert kwargs["status"] == "error"
-    assert any("decrypt" in e.lower() for e in kwargs["errors"])
-
-
-@pytest.mark.asyncio
-async def test_run_cloud_agent_happy_path_gmail():
-    """Cloud agent happy path: Gmail fetch → LLM extraction → inserts → success."""
-    from app.integrations import EmailMessage, encrypt_token
-    from cryptography.fernet import Fernet as _Fernet
-
-    fernet_key = _Fernet.generate_key().decode()
-    credentials = {
-        "token": "access_abc",
-        "refresh_token": "refresh_xyz",
-        "token_uri": "https://oauth2.googleapis.com/token",
-        "client_id": "cid",
-        "client_secret": "csec",
-    }
-
-    config = _make_cloud_config()
-    config.provider = "gmail"
-    config.prompt_template = "Extract tasks from this email."
-    config.data_types = ["tasks"]
-
-    with patch("app.integrations.settings") as ms:
-        ms.OAUTH_ENCRYPTION_KEY = fernet_key
-        config.oauth_token_encrypted = encrypt_token(credentials)
-
-    run_log = _make_run_log(config.id, agent_type="cloud")
-    mgr = _make_manager()
-
-    sample_email = EmailMessage(
-        id="msg001",
-        subject="Action required",
-        sender="boss@company.com",
-        body_text="Please fix the bug by Friday.",
-        date=datetime(2025, 6, 1, 10, 0, tzinfo=timezone.utc),
-    )
-
-    extracted_items = [{"table": "tasks", "data": {"title": "Fix the bug", "priority": "high"}}]
-
-    with patch("app.integrations.settings") as mock_int_settings, \
-         patch("app.core.agent_runner._finalize_run", new_callable=AsyncMock) as mock_finalize, \
-         patch("app.core.agent_runner._extract_items_from_content", new_callable=AsyncMock, return_value=extracted_items) as mock_extract, \
-         patch("app.core.agent_runner._send_insert_to_client", new_callable=AsyncMock, return_value={"ok": True}) as mock_insert, \
-         patch("app.core.agent_runner.async_session"):
-        mock_int_settings.OAUTH_ENCRYPTION_KEY = fernet_key
-
-        mock_gmail = AsyncMock()
-        mock_gmail.fetch_messages = AsyncMock(return_value=[sample_email])
-        mock_gmail.refreshed_credentials = None
-
-        with patch("app.integrations.decrypt_token", return_value=credentials), \
-             patch("app.integrations.get_provider", return_value=mock_gmail):
-            await run_cloud_agent(_FREE_UID, config, run_log, mgr)
-
-    mock_extract.assert_called_once()
-    mock_insert.assert_called_once()
-    _, kwargs = mock_finalize.call_args
-    assert kwargs["status"] == "success"
-    assert kwargs["items_processed"] == 1
-    assert kwargs["items_created"] == 1
-    assert kwargs["config_type"] == "cloud"
-
-
-@pytest.mark.asyncio
-async def test_run_cloud_agent_provider_fetch_error():
-    """Cloud agent records error status when provider fetch raises RuntimeError."""
-    credentials = {"token": "abc"}
-    config = _make_cloud_config()
-    config.oauth_token_encrypted = "some_encrypted_value"  # non-empty so decrypt step is reached
-    config.prompt_template = "Extract tasks."
-    config.data_types = ["tasks"]
-    run_log = _make_run_log(config.id, agent_type="cloud")
-    mgr = _make_manager()
-
-    mock_provider = AsyncMock()
-    mock_provider.fetch_messages = AsyncMock(side_effect=RuntimeError("API quota exceeded"))
-    mock_provider.refreshed_credentials = None
-
-    with patch("app.core.agent_runner._finalize_run", new_callable=AsyncMock) as mock_finalize, \
-         patch("app.integrations.decrypt_token", return_value=credentials), \
-         patch("app.integrations.get_provider", return_value=mock_provider), \
-         patch("app.core.agent_runner.async_session"):
-        await run_cloud_agent(_FREE_UID, config, run_log, mgr)
-
-    _, kwargs = mock_finalize.call_args
-    assert kwargs["status"] == "error"
-    assert any("quota" in e.lower() or "fetch" in e.lower() for e in kwargs["errors"])
-
-
-@pytest.mark.asyncio
-async def test_run_cloud_agent_refreshed_token_persisted():
-    """When the provider refreshes its token, the new ciphertext is written to DB."""
-    from app.integrations import EmailMessage, encrypt_token
-    from cryptography.fernet import Fernet as _Fernet
-
-    fernet_key = _Fernet.generate_key().decode()
-    credentials = {"token": "old_token", "refresh_token": "rt_old"}
-    fresh_credentials = {"token": "new_token", "refresh_token": "rt_new"}
-
-    config = _make_cloud_config()
-    config.prompt_template = "Extract tasks."
-    config.data_types = ["tasks"]
-
-    with patch("app.integrations.settings") as ms:
-        ms.OAUTH_ENCRYPTION_KEY = fernet_key
-        config.oauth_token_encrypted = encrypt_token(credentials)
-
-    run_log = _make_run_log(config.id, agent_type="cloud")
-    mgr = _make_manager()
-
-    mock_provider = AsyncMock()
-    mock_provider.fetch_messages = AsyncMock(return_value=[])
-    mock_provider.refreshed_credentials = fresh_credentials  # token was refreshed
-
-    # Track DB writes via mock async_session.
-    mock_cfg_row = MagicMock()
-    mock_cfg_row.oauth_token_encrypted = None
-
-    mock_db = AsyncMock()
-    mock_db.__aenter__ = AsyncMock(return_value=mock_db)
-    mock_db.__aexit__ = AsyncMock(return_value=False)
-    mock_db.scalar_one_or_none = AsyncMock(return_value=mock_cfg_row)
-    cfg_result = MagicMock()
-    cfg_result.scalar_one_or_none.return_value = mock_cfg_row
-    mock_db.execute = AsyncMock(return_value=cfg_result)
-    mock_db.commit = AsyncMock()
-
-    with patch("app.core.agent_runner._finalize_run", new_callable=AsyncMock), \
-         patch("app.integrations.decrypt_token", return_value=credentials), \
-         patch("app.integrations.get_provider", return_value=mock_provider), \
-         patch("app.integrations.encrypt_token", return_value="new_encrypted") as mock_encrypt, \
-         patch("app.core.agent_runner.async_session", return_value=mock_db), \
-         patch("app.integrations.settings") as mock_int_settings:
-        mock_int_settings.OAUTH_ENCRYPTION_KEY = fernet_key
-        await run_cloud_agent(_FREE_UID, config, run_log, mgr)
-
-    # The new encrypted token should have been written to the config row.
-    mock_encrypt.assert_called_once_with(fresh_credentials)
-    assert mock_cfg_row.oauth_token_encrypted == "new_encrypted"
-
-
-@pytest.mark.asyncio
-async def test_finalize_run_updates_cloud_config_last_run_at():
-    """_finalize_run with config_type='cloud' updates CloudAgentConfig.last_run_at."""
-    from app.core.agent_runner import _finalize_run
-
-    run_log = _make_run_log(str(uuid.uuid4()), agent_type="cloud")
-    run_log.id = str(uuid.uuid4())
-
-    mock_cfg = MagicMock()
-    mock_cfg.last_run_at = None
-
-    cfg_result = MagicMock()
-    cfg_result.scalar_one_or_none.return_value = mock_cfg
-
-    mock_db = AsyncMock()
-    mock_db.__aenter__ = AsyncMock(return_value=mock_db)
-    mock_db.__aexit__ = AsyncMock(return_value=False)
-    mock_db.merge = AsyncMock(return_value=run_log)
-    mock_db.execute = AsyncMock(return_value=cfg_result)
-    mock_db.commit = AsyncMock()
-
-    config_id = str(uuid.uuid4())
-
-    with patch("app.core.agent_runner.async_session", return_value=mock_db):
-        await _finalize_run(
-            run_log,
-            status="success",
-            update_config_last_run=True,
-            config_id=config_id,
-            config_type="cloud",
-        )
-
-    # CloudAgentConfig.last_run_at should have been set.
-    assert mock_cfg.last_run_at is not None
-    mock_db.commit.assert_called()
-
-
-# ---------------------------------------------------------------------------
-# trigger_pending_runs
-# ---------------------------------------------------------------------------
-
-
-@pytest.mark.asyncio
-async def test_trigger_pending_runs_no_overdue():
-    """Pending-run scan is skipped because agent config is client-owned."""
-
-    mgr = _make_manager()
-
-    with patch("app.core.agent_runner.run_local_agent", new_callable=AsyncMock) as mock_run:
-        await trigger_pending_runs(_FREE_UID, "dev-001", mgr)
-
-    mock_run.assert_not_called()
-
-
-@pytest.mark.asyncio
-async def test_trigger_pending_runs_device_id_filter():
-    """Device filtering is no longer backend-managed in pending runs."""
-
-    mgr = _make_manager(device_id="dev-001")
-
-    with patch("app.core.agent_runner.run_local_agent", new_callable=AsyncMock) as mock_run:
-        await trigger_pending_runs(_FREE_UID, "dev-001", mgr)
-
-    mock_run.assert_not_called()
-
-
-@pytest.mark.asyncio
-async def test_trigger_pending_runs_dispatches_overdue():
-    """No pending runs are dispatched by backend after config deprecation."""
-
-    mgr = _make_manager()
-
-    with patch("app.core.agent_runner.run_local_agent", new_callable=AsyncMock) as mock_run:
-        await trigger_pending_runs(_FREE_UID, "dev-001", mgr)
-
-    mock_run.assert_not_called()
-
-
-# ---------------------------------------------------------------------------
-# Integration: POST /agents/can-create and /agents/trigger
-# ---------------------------------------------------------------------------
-
-
-@pytest.fixture(autouse=True)
-def _override_db(db_session):
-    """Route all get_session calls to the test SQLite session."""
-
-    async def _gen():
-        yield db_session
-
-    app.dependency_overrides[get_session] = _gen
-    yield
-    app.dependency_overrides.pop(get_session, None)
-
-
-@pytest.mark.asyncio
-async def test_can_create_agent_allows_when_under_limit(client):
-    """POST /agents/can-create returns allowed=True when under tier limit."""
-    resp = client.post(
-        "/api/v1/agents/can-create",
-        json={"active_agents": 0},
-        headers=auth_header("free"),
-    )
-    assert resp.status_code == 200
-    body = resp.json()
-    assert body["allowed"] is True
-    assert body["tier"] == "free"
-    assert body["active_agents"] == 0
-    assert body["limit"] == 2
-
-
-@pytest.mark.asyncio
-async def test_can_create_agent_denies_when_at_limit(client):
-    """POST /agents/can-create returns allowed=False at free-tier limit."""
-    resp = client.post(
-        "/api/v1/agents/can-create",
-        json={"active_agents": 2},
-        headers=auth_header("free"),
-    )
-    assert resp.status_code == 200
-    body = resp.json()
-    assert body["allowed"] is False
-    assert body["limit"] == 2
-
-
-@pytest.mark.asyncio
-async def test_trigger_run_local_agent_creates_run_log(client, db_session):
-    """POST /agents/trigger creates a local run log and dispatches background task."""
-    dispatched: list[tuple[str, str]] = []
-
-    async def _fake_run(user_id, cfg, run_log, device_mgr):
-        dispatched.append((user_id, cfg.id))
-
-    def _fake_create_task(coro):
-        coro.close()
-        return MagicMock()
-
-    with patch("app.api.routes.agents.run_local_agent", new_callable=AsyncMock, side_effect=_fake_run), \
-         patch("asyncio.create_task") as mock_create_task:
-        mock_create_task.side_effect = _fake_create_task
-        resp = client.post(
-            "/api/v1/agents/trigger",
-            json={
-                "directory": "/home/user/docs",
-                "what_to_extract": ["task", "note"],
-                "actions_by_type": {"task": ["add", "update"], "note": ["add"]},
-                "batch_interval": "0 */6 * * *",
-                "custom_agent_prompt": "Extract tasks and notes.",
-                "active_agents": 0,
-            },
-            headers=auth_header("power"),
-        )
-
-    assert resp.status_code == 202
-    data = resp.json()
-    assert isinstance(data["agent_id"], str)
-    assert data["agent_id"]
-    assert data["status"] == "running"
-    assert data["agent_type"] == "local"
-
-    # Verify create_task was called (dispatching background run).
-    mock_create_task.assert_called_once()

tests/test_agent_runner_v2.py

@@ -0,0 +1,431 @@
+"""Tests for Local Agent V2 runner (Step 2).
+
+Covers the unified per-file flow:
+  Phase A — detect + preprocess (Python, zero LLM)
+  Phase B — single LLM call with tools (classify + extract + create)
+
+Fixture-based eval tests (2.1–2.7)
+-----------------------------------
+Cases are defined in tests/fixtures/agent_runner_v2/cases.yaml.
+Email HTML files live in tests/fixtures/agent_runner_v2/data/.
+Use --runner-dir to point at a custom folder (same structure required).
+
+Unit tests (no LLM)
+--------------------
+  2.8  items_created count   → items_created == N create_* calls
+  2.9  Device offline        → status=error
+  2.10 Empty file            → items_processed=0, status=success
+
+Run:
+    pytest tests/test_agent_runner_v2.py -v
+    pytest tests/test_agent_runner_v2.py -v -k "2_9 or 2_10 or 2_8"   # unit only
+    pytest tests/test_agent_runner_v2.py -v -k "eval"                  # LLM evals only
+    pytest tests/test_agent_runner_v2.py -v --runner-dir /path/to/dir  # custom fixtures
+"""
+
+from __future__ import annotations
+
+import uuid
+from contextlib import nullcontext
+from datetime import datetime, timezone
+from pathlib import Path
+from typing import Any
+from unittest.mock import AsyncMock, MagicMock, patch
+
+import pytest
+import yaml
+
+from app.core.agent_runner import (
+    _format_metadata,
+    _format_projects,
+    _get_extraction_rules,
+    _get_no_match_behavior,
+    run_local_agent,
+)
+from app.core.device_manager import DeviceConnectionManager
+from app.core.langfuse_client import get_langfuse
+from app.models import AgentRunLog, LocalAgentConfig
+from tests.conftest import TEST_USER_IDS
+
+# ── Constants ─────────────────────────────────────────────────────────────
+
+_USER_ID = TEST_USER_IDS["power"]
+
+_DEFAULT_FIXTURE_DIR = Path(__file__).parent / "fixtures" / "agent_runner_v2"
+
+_AGENT_CONFIG = {
+    "content_types": [
+        {
+            "id": "email_html",
+            "label": "Email HTML",
+            "detection_hint": "HTML file with From/To/Subject headers",
+            "preprocessing": "email_html",
+            "extraction_prompt": (
+                "If the email contains a direct action request or task assignment → create a task. "
+                "If the email contains informational content, updates, or FYI → create a note. "
+                "If the email mentions a specific date for a meeting or deadline → create a timeline entry."
+            ),
+        }
+    ],
+    "global_rules": [
+        "Se il file non è riconducibile a nessun progetto, non creare alcuna entità."
+    ],
+    "data_types": ["tasks", "notes", "timelines"],
+}
+
+# Canonical project definitions, referenced symbolically in cases.yaml.
+_PROJECTS: dict[str, dict] = {
+    "alpha": {"id": "proj-alpha", "name": "Project Alpha", "status": "active"},
+    "beta":  {"id": "proj-beta",  "name": "Project Beta",  "status": "active"},
+}
+
+
+# ── Fixture loading ───────────────────────────────────────────────────────
+
+
+def _fixtures_dir(config) -> Path:
+    override = config.getoption("--runner-dir")
+    return Path(override) if override else _DEFAULT_FIXTURE_DIR
+
+
+def _load_cases(config) -> list[dict]:
+    return yaml.safe_load(
+        (_fixtures_dir(config) / "cases.yaml").read_text(encoding="utf-8")
+    )
+
+
+def _read_case_file(case: dict, data_dir: Path) -> str:
+    return (data_dir / case["file"]).read_text(encoding="utf-8")
+
+
+def _resolve_projects(entries: list[str | dict]) -> list[dict]:
+    """Resolve project list from YAML: symbolic names and/or inline dicts."""
+    result = []
+    for entry in entries:
+        if isinstance(entry, str):
+            if entry in _PROJECTS:
+                result.append(_PROJECTS[entry])
+        elif isinstance(entry, dict):
+            result.append(entry)
+    return result
+
+
+# ── pytest_generate_tests — parametrize eval tests from YAML ─────────────
+
+
+def pytest_generate_tests(metafunc):
+    if "runner_case" not in metafunc.fixturenames:
+        return
+    cases = _load_cases(metafunc.config)
+    metafunc.parametrize("runner_case", cases, ids=[c["id"] for c in cases])
+
+
+# ── Test helpers ──────────────────────────────────────────────────────────
+
+
+def _make_config(
+    agent_config: dict | None = None,
+    directory: str = "/emails",
+    device_id: str = "dev-001",
+) -> LocalAgentConfig:
+    return LocalAgentConfig(
+        id=str(uuid.uuid4()),
+        user_id=_USER_ID,
+        device_id=device_id,
+        name="Test V2 Agent",
+        directory_paths=[directory],
+        data_types=["tasks", "notes", "timelines"],
+        prompt_template="",
+        agent_config=agent_config or _AGENT_CONFIG,
+        file_extensions=[".html", ".eml"],
+        schedule_cron="0 */6 * * *",
+        enabled=True,
+        last_run_at=None,
+    )
+
+
+def _make_run_log(agent_id: str) -> AgentRunLog:
+    return AgentRunLog(
+        id=str(uuid.uuid4()),
+        agent_id=agent_id,
+        agent_type="local",
+        user_id=_USER_ID,
+        status="running",
+        started_at=datetime.now(timezone.utc),
+    )
+
+
+def _make_manager(online: bool = True) -> DeviceConnectionManager:
+    mgr = DeviceConnectionManager()
+    if online:
+        ws = MagicMock()
+        ws.send_text = AsyncMock()
+        mgr.register(_USER_ID, "dev-001", ws)
+    return mgr
+
+
+def _make_executor(
+    file_path: str,
+    file_content: str,
+    projects: list[dict] | None = None,
+    existing_tasks: list[dict] | None = None,
+    existing_notes: list[dict] | None = None,
+    existing_timelines: list[dict] | None = None,
+) -> tuple[Any, list[dict]]:
+    """Return (async_executor, captured_calls).
+
+    The executor handles all ``execute_on_client`` payloads:
+    directory listing, file reading, project/entity fetching, and CRUD.
+    """
+    calls: list[dict] = []
+    _projects = projects if projects is not None else list(_PROJECTS.values())
+
+    async def _executor(payload: dict) -> dict:
+        action = payload.get("action", "")
+        table = payload.get("table", "")
+        data = payload.get("data") or {}
+        calls.append({"action": action, "table": table, "data": data})
+
+        if action == "list_directory":
+            return {"entries": [{"type": "file", "path": file_path}]}
+
+        if action == "get_file_metadata":
+            return {"modifiedAt": None}
+
+        if action == "read_file_content":
+            return {"content": file_content}
+
+        if action == "select":
+            if table == "projects":
+                return {"rows": _projects}
+            if table == "tasks":
+                return {"rows": existing_tasks or []}
+            if table == "notes":
+                return {"rows": existing_notes or []}
+            if table == "timelines":
+                return {"rows": existing_timelines or []}
+            return {"rows": []}
+
+        if action == "insert":
+            return {"row": {"id": str(uuid.uuid4()), **data}}
+
+        if action == "update":
+            return {"success": True}
+
+        return {}
+
+    return _executor, calls
+
+
+# ── Unit: helper functions ────────────────────────────────────────────────
+
+
+def test_format_projects_empty():
+    assert "(no projects" in _format_projects([])
+
+
+def test_format_projects_with_data():
+    result = _format_projects([_PROJECTS["alpha"]])
+    assert "proj-alpha" in result
+    assert "Project Alpha" in result
+
+
+def test_format_metadata_empty():
+    assert _format_metadata({}) == ""
+
+
+def test_format_metadata_email():
+    meta = {"subject": "Fix bug", "from": "boss@co.com", "date": "2026-04-07"}
+    result = _format_metadata(meta)
+    assert "Fix bug" in result
+    assert "boss@co.com" in result
+
+
+def test_get_extraction_rules_match():
+    rules = _get_extraction_rules(_AGENT_CONFIG, "email_html")
+    assert "task" in rules.lower()
+
+
+def test_get_extraction_rules_fallback():
+    rules = _get_extraction_rules(_AGENT_CONFIG, "plain_text")
+    assert "extract" in rules.lower()
+
+
+def test_get_no_match_behavior_from_global_rules():
+    behavior = _get_no_match_behavior(_AGENT_CONFIG)
+    assert behavior  # non-empty
+
+
+def test_get_no_match_behavior_default():
+    behavior = _get_no_match_behavior({})
+    assert "project" in behavior.lower()
+
+
+# ── Unit: 2.9 — device offline ───────────────────────────────────────────
+
+
+@pytest.mark.asyncio
+async def test_2_9_device_offline():
+    """2.9 No device online → status=error, no executor created."""
+    config = _make_config()
+    run_log = _make_run_log(config.id)
+    mgr = _make_manager(online=False)
+
+    with patch("app.core.agent_runner._finalize_run", new_callable=AsyncMock) as mock_fin:
+        await run_local_agent(_USER_ID, config, run_log, mgr)
+
+    _, kwargs = mock_fin.call_args
+    assert kwargs["status"] == "error"
+    assert any("not connected" in e for e in kwargs.get("errors", []))
+
+
+# ── Unit: 2.10 — empty file ──────────────────────────────────────────────
+
+
+@pytest.mark.asyncio
+async def test_2_10_empty_file():
+    """2.10 File with empty content → skipped, items_processed=0, success."""
+    config = _make_config()
+    run_log = _make_run_log(config.id)
+    mgr = _make_manager()
+
+    executor, calls = _make_executor(
+        file_path="/emails/empty.html",
+        file_content="",
+        projects=[_PROJECTS["alpha"]],
+    )
+
+    with patch("app.core.agent_runner._make_agent_executor", return_value=executor), \
+         patch("app.core.agent_runner._finalize_run", new_callable=AsyncMock) as mock_fin:
+        await run_local_agent(_USER_ID, config, run_log, mgr)
+
+    _, kwargs = mock_fin.call_args
+    assert kwargs["items_processed"] == 0
+    assert kwargs["status"] == "success"
+    assert kwargs["items_created"] == 0
+
+
+# ── Unit: 2.8 — items_created count ─────────────────────────────────────
+
+
+@pytest.mark.asyncio
+async def test_2_8_items_created_count():
+    """2.8 items_created == number of create_* tool calls per run."""
+    config = _make_config()
+    run_log = _make_run_log(config.id)
+    mgr = _make_manager()
+
+    executor, _calls = _make_executor(
+        file_path="/emails/action.html",
+        file_content="<html><body><p>Fix the login bug in Project Alpha.</p></body></html>",
+        projects=[_PROJECTS["alpha"]],
+    )
+
+    async def mock_run_agent(*, _tool_calls_out=None, **kw) -> str:
+        if _tool_calls_out is not None:
+            _tool_calls_out.extend(["create_task", "create_note", "update_task"])
+        return "Done."
+
+    with patch("app.core.agent_runner._make_agent_executor", return_value=executor), \
+         patch("app.core.agent_runner._run_agent_with_tools", side_effect=mock_run_agent), \
+         patch("app.core.agent_runner._finalize_run", new_callable=AsyncMock) as mock_fin:
+        await run_local_agent(_USER_ID, config, run_log, mgr)
+
+    _, kwargs = mock_fin.call_args
+    # Only create_task + create_note count (not update_task).
+    assert kwargs["items_created"] == 2
+    assert kwargs["items_processed"] == 1
+
+
+# ── Eval: 2.1–2.7 — fixture-driven, real LLM + Langfuse scoring ──────────
+#
+# Cases loaded from tests/fixtures/agent_runner_v2/cases.yaml.
+# Supported assertions (from YAML):
+#   expect_insert: <table>   → at least 1 insert in that table
+#   expect_no_insert: true   → zero inserts in any table
+#   expect_project_id: <id>  → any insert carries this projectId
+#   expect_dedup: true       → task inserts == 0 OR task updates >= 1
+# ─────────────────────────────────────────────────────────────────────────
+
+
+@pytest.mark.asyncio
+@pytest.mark.eval
+async def test_eval_runner(runner_case, pytestconfig):
+    """Parametrized eval test — one invocation per YAML case."""
+    case: dict = runner_case
+    data_dir = _fixtures_dir(pytestconfig) / "data"
+    file_content = _read_case_file(case, data_dir)
+    projects = _resolve_projects(case.get("projects", []))
+
+    config = _make_config()
+    run_log = _make_run_log(config.id)
+    mgr = _make_manager()
+
+    executor, calls = _make_executor(
+        file_path=case["file_path"],
+        file_content=file_content,
+        projects=projects,
+        existing_tasks=case.get("existing_tasks"),
+        existing_notes=case.get("existing_notes"),
+        existing_timelines=case.get("existing_timelines"),
+    )
+
+    lf = get_langfuse()
+    obs_ctx = lf.start_as_current_observation(
+        name=f"eval-runner-{case['id']}-{case.get('score_name', 'unknown').replace('.', '-')}",
+        metadata={"step": "2", "case_id": case["id"]},
+    ) if lf else nullcontext()
+
+    with obs_ctx as obs:
+        with patch("app.core.agent_runner._make_agent_executor", return_value=executor), \
+             patch("app.core.agent_runner._finalize_run", new_callable=AsyncMock) as mock_fin:
+            await run_local_agent(_USER_ID, config, run_log, mgr)
+
+        _, kwargs = mock_fin.call_args
+        inserts = [c for c in calls if c["action"] == "insert"]
+        score, comment = _evaluate_case(case, calls, kwargs)
+
+        if obs is not None:
+            obs.score(
+                name=case.get("score_name", f"runner.case_{case['id']}"),
+                value=score,
+                comment=comment,
+            )
+
+    if lf:
+        lf.flush()
+
+    assert score == 1.0, f"[{case['id']}] {case.get('description', '')} — {comment}"
+
+
+def _evaluate_case(case: dict, calls: list[dict], finalize_kwargs: dict) -> tuple[float, str]:
+    """Return (score, comment) for a YAML case given the captured executor calls."""
+    inserts = [c for c in calls if c["action"] == "insert"]
+
+    if case.get("expect_no_insert"):
+        score = 1.0 if len(inserts) == 0 else 0.0
+        return score, f"inserts={len(inserts)} (expected 0)"
+
+    if "expect_insert" in case:
+        tables = case["expect_insert"]
+        if isinstance(tables, str):
+            tables = [tables]
+        missing = [t for t in tables if not any(c["table"] == t for c in inserts)]
+        score = 1.0 if not missing else 0.0
+        counts = {t: sum(1 for c in inserts if c["table"] == t) for t in tables}
+        return score, f"inserts={counts}" + (f" missing={missing}" if missing else "")
+
+    if "expect_project_id" in case:
+        expected_pid = case["expect_project_id"]
+        correct = any(c.get("data", {}).get("projectId") == expected_pid for c in inserts)
+        score = 1.0 if correct else 0.0
+        all_pids = [c.get("data", {}).get("projectId") for c in inserts]
+        return score, f"projectIds={all_pids} (expected {expected_pid!r})"
+
+    if case.get("expect_dedup"):
+        task_creates = [c for c in inserts if c["table"] == "tasks"]
+        task_updates = [c for c in calls if c["action"] == "update" and c["table"] == "tasks"]
+        score = 1.0 if len(task_creates) == 0 or len(task_updates) >= 1 else 0.0
+        return score, f"task_creates={len(task_creates)} task_updates={len(task_updates)}"
+
+    return 0.0, "no assertion defined in case"

tests/test_agent_setup.py

@@ -1,243 +0,0 @@
-"""Tests for the Chatbot Journey endpoints.
-
-Covers:
-  1. Start journey for local agent → session_id + first question, done=False
-  2. Start journey for cloud agent → contextual email-focused question
-  3. Start journey with existing agent_id → session seeded, first question returned
-  4. Start journey with non-existent agent_id → still succeeds (graceful fallback)
-  5. Message: continue conversation → done=False, follow-up question returned
-  6. Message: LLM wraps up → done=True + prompt_template extracted correctly
-  7. Message with max-turns nudge → no crash, returns response
-  8. Invalid session_id → 404
-  9. Expired session → 404
-  10. Session ownership: user B cannot access user A's session
-  11. No JWT on /start → 401
-  12. No JWT on /message → 401
-"""
-
-from __future__ import annotations
-
-import time
-import uuid
-from unittest.mock import AsyncMock, patch
-
-import pytest
-from fastapi.testclient import TestClient
-from sqlalchemy.ext.asyncio import AsyncSession
-
-from app.api.routes.agent_setup import (
-    _SESSION_TTL_SECONDS,
-    _TEMPLATE_END,
-    _TEMPLATE_START,
-    _extract_template,
-    _sessions,
-)
-from app.models import LocalAgentConfig
-from tests.conftest import TEST_USER_IDS, auth_header
-
-# ── Helpers ──────────────────────────────────────────────────────────────
-
-
-def _start(client: TestClient, agent_type: str = "local", agent_id: str | None = None, tier: str = "power") -> dict:
-    body: dict = {"agent_type": agent_type}
-    if agent_id:
-        body["agent_id"] = agent_id
-    resp = client.post("/api/v1/agents/journey/start", json=body, headers=auth_header(tier))
-    return resp
-
-
-def _message(client: TestClient, session_id: str, message: str, tier: str = "power") -> dict:
-    return client.post(
-        "/api/v1/agents/journey/message",
-        json={"session_id": session_id, "message": message},
-        headers=auth_header(tier),
-    )
-
-
-# ── Unit: _extract_template ───────────────────────────────────────────────
-
-
-def test_extract_template_present():
-    text = f"Some preamble.\n{_TEMPLATE_START}\nExtract tasks from emails.\n{_TEMPLATE_END}\nTrailing text."
-    result = _extract_template(text)
-    assert result == "Extract tasks from emails."
-
-
-def test_extract_template_absent():
-    assert _extract_template("No markers here.") is None
-
-
-def test_extract_template_empty_content():
-    text = f"{_TEMPLATE_START}\n{_TEMPLATE_END}"
-    assert _extract_template(text) is None
-
-
-# ── Start journey ─────────────────────────────────────────────────────────
-
-
-def test_start_journey_local(client: TestClient):
-    resp = _start(client, agent_type="local")
-    assert resp.status_code == 200
-    body = resp.json()
-    assert "session_id" in body
-    assert body["done"] is False
-    assert body["prompt_template"] is None
-    assert len(body["message"]) > 0
-    # Local question should be about files/directories
-    assert any(w in body["message"].lower() for w in ("file", "director", "document", "monitor"))
-
-
-def test_start_journey_cloud(client: TestClient):
-    resp = _start(client, agent_type="cloud")
-    assert resp.status_code == 200
-    body = resp.json()
-    assert body["done"] is False
-    # Cloud question should mention emails or messages
-    assert any(w in body["message"].lower() for w in ("email", "message", "communication"))
-
-
-def test_start_journey_with_agent_id(client: TestClient, db_session: AsyncSession):
-    """When agent_id is provided, session should be created even if agent doesn't exist."""
-    fake_agent_id = str(uuid.uuid4())
-    resp = _start(client, agent_type="local", agent_id=fake_agent_id)
-    # Should succeed gracefully even if the agent_id doesn't exist
-    assert resp.status_code == 200
-    body = resp.json()
-    assert body["done"] is False
-
-
-def test_start_journey_with_existing_agent(client: TestClient, db_session: AsyncSession):
-    """When a real local agent is provided, session is seeded with its prompt_template."""
-    import asyncio
-
-    user_id = TEST_USER_IDS["power"]
-    agent = LocalAgentConfig(
-        id=str(uuid.uuid4()),
-        user_id=user_id,
-        name="Test Agent",
-        device_id="device-1",
-        directory_paths=["/home/user/emails"],
-        data_types=["tasks"],
-        prompt_template="Extract tasks from .eml files.",
-        file_extensions=[".eml"],
-        schedule_cron="0 */6 * * *",
-        enabled=True,
-    )
-
-    async def _seed():
-        db_session.add(agent)
-        await db_session.commit()
-
-    asyncio.get_event_loop().run_until_complete(_seed())
-
-    resp = _start(client, agent_type="local", agent_id=agent.id)
-    assert resp.status_code == 200
-    body = resp.json()
-    assert body["done"] is False
-    # The session should be stored
-    assert body["session_id"] in _sessions
-
-
-def test_start_journey_requires_auth(client: TestClient):
-    resp = client.post("/api/v1/agents/journey/start", json={"agent_type": "local"})
-    assert resp.status_code == 401
-
-
-# ── Message ───────────────────────────────────────────────────────────────
-
-
-def test_message_continues_conversation(client: TestClient):
-    """A mid-journey reply (no template markers) returns done=False."""
-    follow_up = "That looks good. Can you tell me more about priority rules?"
-
-    with patch("app.api.routes.agent_setup._call_llm", new=AsyncMock(return_value=follow_up)):
-        start_resp = _start(client, agent_type="local")
-        assert start_resp.status_code == 200
-        session_id = start_resp.json()["session_id"]
-
-        msg_resp = _message(client, session_id, "I have .eml and .txt files")
-        assert msg_resp.status_code == 200
-        body = msg_resp.json()
-        assert body["done"] is False
-        assert body["prompt_template"] is None
-        assert body["message"] == follow_up
-        assert body["session_id"] == session_id
-
-
-def test_message_produces_template(client: TestClient):
-    """When the LLM includes PROMPT_TEMPLATE markers, done=True and prompt_template is set."""
-    final_template = "Extract tasks from email. Subject → title. 'urgent' → high priority."
-    llm_response = (
-        "Great, I have all the information I need.\n"
-        f"{_TEMPLATE_START}\n{final_template}\n{_TEMPLATE_END}\n"
-    )
-
-    with patch("app.api.routes.agent_setup._call_llm", new=AsyncMock(return_value=llm_response)):
-        start_resp = _start(client, agent_type="cloud")
-        assert start_resp.status_code == 200
-        session_id = start_resp.json()["session_id"]
-
-        msg_resp = _message(client, session_id, "Only invoices from clients")
-        assert msg_resp.status_code == 200
-        body = msg_resp.json()
-        assert body["done"] is True
-        assert body["prompt_template"] == final_template
-        # Session should be cleaned up
-        assert session_id not in _sessions
-
-
-def test_message_invalid_session(client: TestClient):
-    resp = _message(client, "nonexistent-session-id", "hello")
-    assert resp.status_code == 404
-
-
-def test_message_wrong_owner(client: TestClient):
-    """User B cannot access user A's session."""
-    start_resp = _start(client, agent_type="local", tier="power")
-    session_id = start_resp.json()["session_id"]
-
-    # user with "pro" tier (different user_id) tries to send a message
-    resp = client.post(
-        "/api/v1/agents/journey/message",
-        json={"session_id": session_id, "message": "hello"},
-        headers=auth_header("pro"),  # different user
-    )
-    assert resp.status_code == 404
-
-
-def test_message_expired_session(client: TestClient):
-    """Expired sessions return 404."""
-    start_resp = _start(client, agent_type="local")
-    session_id = start_resp.json()["session_id"]
-
-    # Manually expire the session
-    _sessions[session_id].created_at = time.monotonic() - _SESSION_TTL_SECONDS - 1
-
-    resp = _message(client, session_id, "hello")
-    assert resp.status_code == 404
-
-
-def test_message_requires_auth(client: TestClient):
-    resp = client.post(
-        "/api/v1/agents/journey/message",
-        json={"session_id": "any", "message": "hello"},
-    )
-    assert resp.status_code == 401
-
-
-def test_message_max_turns_nudge(client: TestClient):
-    """After _MAX_TURNS user messages, a system nudge is appended but no crash occurs."""
-    from app.api.routes.agent_setup import _MAX_TURNS
-
-    follow_up = "Tell me more about priority rules."
-
-    with patch("app.api.routes.agent_setup._call_llm", new=AsyncMock(return_value=follow_up)):
-        start_resp = _start(client, agent_type="local")
-        session_id = start_resp.json()["session_id"]
-
-        for i in range(_MAX_TURNS):
-            resp = _message(client, session_id, f"Answer {i + 1}")
-            assert resp.status_code == 200
-            # While no template produced, session must still exist
-            if resp.json()["done"]:
-                break  # LLM decided to wrap up early — also fine

tests/test_auth.py

@@ -1,4 +1,4 @@
-"""Tests for auth routes: register, login, refresh, me.
+"""Tests for auth routes: register, login, refresh, me, OAuth social login.
 
 Exercises the full auth lifecycle through the FastAPI TestClient against the
 in-memory SQLite test database seeded by ``conftest.py``.
@@ -7,9 +7,11 @@ in-memory SQLite test database seeded by ``conftest.py``.
 from __future__ import annotations
 
 import time
+from unittest.mock import AsyncMock, patch
 
 from jose import jwt
 
+from app.auth.oauth_providers import GoogleOAuthProvider, OAuthUserInfo
 from app.config.settings import settings
 from tests.conftest import auth_header, TEST_USER_IDS
 
@@ -204,3 +206,153 @@ class TestMe:
         token = jwt.encode(payload, "wrong-secret", algorithm="HS256")
         resp = client.get("/api/v1/auth/me", headers={"Authorization": f"Bearer {token}"})
         assert resp.status_code == 401
+
+
+# ── TestOAuth ─────────────────────────────────────────────────────────
+
+
+class TestOAuth:
+    """GET /auth/oauth/google/authorize and POST /auth/oauth/google/callback."""
+
+    FAKE_PROVIDER_USER_ID = "google-sub-12345"
+    FAKE_EMAIL = "oauth@example.com"
+    FAKE_AVATAR = "https://lh3.googleusercontent.com/photo.jpg"
+
+    def _patch_google(self, monkeypatch) -> None:
+        monkeypatch.setattr(settings, "GOOGLE_AUTH_CLIENT_ID", "fake-client-id")
+        monkeypatch.setattr(settings, "GOOGLE_AUTH_CLIENT_SECRET", "fake-client-secret")
+
+    def _userinfo(
+        self,
+        email: str | None = None,
+        email_verified: bool = True,
+    ) -> OAuthUserInfo:
+        return OAuthUserInfo(
+            provider_user_id=self.FAKE_PROVIDER_USER_ID,
+            email=email or self.FAKE_EMAIL,
+            email_verified=email_verified,
+            avatar_url=self.FAKE_AVATAR,
+            name="OAuth User",
+        )
+
+    def _authorize(self, client) -> str:
+        """Call /authorize and return the fresh state token."""
+        resp = client.get("/api/v1/auth/oauth/google/authorize")
+        assert resp.status_code == 200
+        return resp.json()["state"]
+
+    def _callback(self, client, state: str, userinfo: OAuthUserInfo):
+        """POST /callback with mocked provider exchange_code + get_userinfo."""
+        with (
+            patch.object(
+                GoogleOAuthProvider,
+                "exchange_code",
+                new=AsyncMock(return_value={"access_token": "google-access-tok"}),
+            ),
+            patch.object(
+                GoogleOAuthProvider,
+                "get_userinfo",
+                new=AsyncMock(return_value=userinfo),
+            ),
+        ):
+            return client.post(
+                "/api/v1/auth/oauth/google/callback",
+                json={"code": "auth-code", "state": state},
+            )
+
+    def _decode_sub(self, access_token: str) -> str:
+        return jwt.decode(
+            access_token, settings.JWT_SECRET, algorithms=[settings.JWT_ALGORITHM]
+        )["sub"]
+
+    # -- authorize --
+
+    def test_authorize_returns_url_and_state(self, client, monkeypatch) -> None:
+        self._patch_google(monkeypatch)
+        resp = client.get("/api/v1/auth/oauth/google/authorize")
+        assert resp.status_code == 200
+        data = resp.json()
+        assert "url" in data and "state" in data
+        assert "accounts.google.com" in data["url"]
+        assert len(data["state"]) > 0
+
+    def test_authorize_unconfigured_returns_503(self, client, monkeypatch) -> None:
+        monkeypatch.setattr(settings, "GOOGLE_AUTH_CLIENT_ID", "")
+        monkeypatch.setattr(settings, "GOOGLE_AUTH_CLIENT_SECRET", "")
+        resp = client.get("/api/v1/auth/oauth/google/authorize")
+        assert resp.status_code == 503
+
+    # -- callback --
+
+    def test_callback_state_mismatch_returns_401(self, client, monkeypatch) -> None:
+        self._patch_google(monkeypatch)
+        resp = client.post(
+            "/api/v1/auth/oauth/google/callback",
+            json={"code": "code", "state": "not-a-real-state"},
+        )
+        assert resp.status_code == 401
+
+    def test_callback_creates_new_user(self, client, monkeypatch) -> None:
+        """First-time Google login creates a new user and returns valid tokens."""
+        self._patch_google(monkeypatch)
+        state = self._authorize(client)
+        resp = self._callback(client, state, self._userinfo())
+
+        assert resp.status_code == 200
+        data = resp.json()
+        assert "access_token" in data and "refresh_token" in data
+        payload = jwt.decode(
+            data["access_token"], settings.JWT_SECRET, algorithms=[settings.JWT_ALGORITHM]
+        )
+        assert payload["email"] == self.FAKE_EMAIL
+
+    def test_callback_existing_oauth_link_logs_in(self, client, monkeypatch) -> None:
+        """Second Google login with the same account re-uses the existing user."""
+        self._patch_google(monkeypatch)
+        userinfo = self._userinfo()
+
+        # First login — creates user + oauth_accounts row
+        resp1 = self._callback(client, self._authorize(client), userinfo)
+        assert resp1.status_code == 200
+        sub1 = self._decode_sub(resp1.json()["access_token"])
+
+        # Second login — finds existing oauth_accounts row → same user
+        resp2 = self._callback(client, self._authorize(client), userinfo)
+        assert resp2.status_code == 200
+        sub2 = self._decode_sub(resp2.json()["access_token"])
+
+        assert sub1 == sub2
+
+    def test_callback_email_match_links_account(self, client, monkeypatch) -> None:
+        """Verified Google email matching an existing password user links the accounts."""
+        email = "link-target@example.com"
+        reg_resp = client.post(
+            "/api/v1/auth/register",
+            json={"email": email, "password": "TestPass123!"},
+        )
+        assert reg_resp.status_code == 201
+        orig_sub = self._decode_sub(reg_resp.json()["access_token"])
+
+        self._patch_google(monkeypatch)
+        state = self._authorize(client)
+        resp = self._callback(client, state, self._userinfo(email=email, email_verified=True))
+
+        assert resp.status_code == 200
+        oauth_sub = self._decode_sub(resp.json()["access_token"])
+        # OAuth login must resolve to the same user as the original registration
+        assert orig_sub == oauth_sub
+
+    def test_callback_unverified_email_conflict_returns_409(self, client, monkeypatch) -> None:
+        """Unverified Google email matching an existing account returns 409, not 500."""
+        email = "conflict@example.com"
+        reg_resp = client.post(
+            "/api/v1/auth/register",
+            json={"email": email, "password": "TestPass123!"},
+        )
+        assert reg_resp.status_code == 201
+
+        self._patch_google(monkeypatch)
+        state = self._authorize(client)
+        resp = self._callback(client, state, self._userinfo(email=email, email_verified=False))
+
+        assert resp.status_code == 409

tests/test_backup.py

@@ -1,243 +0,0 @@
-"""Tests for backup routes: upload, download, history, delete.
-
-Exercises the backup lifecycle through the FastAPI TestClient against the
-in-memory SQLite test database and moto-mocked S3 bucket.
-"""
-
-from __future__ import annotations
-
-import hashlib
-
-
-from tests.conftest import auth_header, TEST_USER_IDS
-
-
-# ── Helpers ───────────────────────────────────────────────────────────
-
-_BLOB = b"encrypted-backup-blob-opaque-bytes"
-_CHECKSUM = hashlib.sha256(_BLOB).hexdigest()
-_VERSION = 1
-_TIMESTAMP = 1700000000000  # arbitrary ms timestamp
-
-
-def _backup_headers(tier: str = "power", **overrides) -> dict[str, str]:
-    """Return auth + backup metadata headers."""
-    headers = auth_header(tier)
-    headers["X-Backup-Version"] = str(overrides.get("version", _VERSION))
-    headers["X-Backup-Timestamp"] = str(overrides.get("timestamp", _TIMESTAMP))
-    headers["X-Backup-Checksum"] = overrides.get("checksum", _CHECKSUM)
-    headers["Content-Type"] = "application/octet-stream"
-    return headers
-
-
-def _upload(client, tier="power", **overrides) -> "Response":  # noqa: F821
-    """Upload a backup blob and return the response."""
-    return client.put(
-        "/api/v1/backup",
-        content=overrides.pop("blob", _BLOB),
-        headers=_backup_headers(tier, **overrides),
-    )
-
-
-# ── TestUploadBackup ──────────────────────────────────────────────────
-
-
-class TestUploadBackup:
-    """PUT /api/v1/backup"""
-
-    def test_upload_success(self, client, s3_bucket) -> None:
-        resp = _upload(client, tier="power")
-        assert resp.status_code == 200
-        assert resp.json() == {"ok": True}
-
-    def test_upload_creates_history_entry(self, client, s3_bucket) -> None:
-        _upload(client, tier="power")
-        history = client.get(
-            "/api/v1/backup/history", headers=auth_header("power")
-        ).json()
-        assert len(history) == 1
-        assert history[0]["version"] == _VERSION
-        assert history[0]["timestamp"] == _TIMESTAMP
-        assert history[0]["checksum"] == _CHECKSUM
-
-    def test_upload_bad_checksum(self, client, s3_bucket) -> None:
-        resp = _upload(client, tier="power", checksum="0" * 64)
-        assert resp.status_code == 400
-
-    def test_upload_free_tier_blocked(self, client, s3_bucket) -> None:
-        """Free tier has backup_gb=0 → should return 402."""
-        resp = _upload(client, tier="free")
-        assert resp.status_code == 402
-
-    def test_upload_pro_tier_allowed(self, client, s3_bucket) -> None:
-        """Pro tier has backup_gb=5 → small blob succeeds."""
-        resp = _upload(client, tier="pro")
-        assert resp.status_code == 200
-
-
-# ── TestDownloadBackup ────────────────────────────────────────────────
-
-
-class TestDownloadBackup:
-    """GET /api/v1/backup"""
-
-    def test_download_latest(self, client, s3_bucket) -> None:
-        _upload(client, tier="power")
-        resp = client.get("/api/v1/backup", headers=auth_header("power"))
-        assert resp.status_code == 200
-        assert resp.content == _BLOB
-        assert resp.headers["X-Checksum"] == _CHECKSUM
-        assert resp.headers["X-Backup-Version"] == str(_VERSION)
-
-    def test_download_no_backup_returns_404(self, client, s3_bucket) -> None:
-        resp = client.get("/api/v1/backup", headers=auth_header("power"))
-        assert resp.status_code == 404
-
-    def test_download_if_modified_since_returns_304(self, client, s3_bucket) -> None:
-        """When If-Modified-Since is after the backup timestamp → 304."""
-        _upload(client, tier="power", timestamp=1700000000000)
-        resp = client.get(
-            "/api/v1/backup",
-            headers={
-                **auth_header("power"),
-                "If-Modified-Since": "Thu, 01 Jan 2099 00:00:00 GMT",
-            },
-        )
-        assert resp.status_code == 304
-
-    def test_download_if_modified_since_returns_200(self, client, s3_bucket) -> None:
-        """When If-Modified-Since is before the backup timestamp → serve blob."""
-        _upload(client, tier="power", timestamp=1700000000000)
-        resp = client.get(
-            "/api/v1/backup",
-            headers={
-                **auth_header("power"),
-                "If-Modified-Since": "Thu, 01 Jan 2000 00:00:00 GMT",
-            },
-        )
-        assert resp.status_code == 200
-        assert resp.content == _BLOB
-
-    def test_download_multiple_returns_latest(self, client, s3_bucket) -> None:
-        """When multiple backups exist, GET returns the one with the highest timestamp."""
-        _upload(client, tier="power", timestamp=1000)
-        blob2 = b"second-encrypted-backup"
-        checksum2 = hashlib.sha256(blob2).hexdigest()
-        _upload(client, tier="power", timestamp=2000, blob=blob2, checksum=checksum2)
-        resp = client.get("/api/v1/backup", headers=auth_header("power"))
-        assert resp.status_code == 200
-        assert resp.content == blob2
-
-
-# ── TestBackupHistory ─────────────────────────────────────────────────
-
-
-class TestBackupHistory:
-    """GET /api/v1/backup/history"""
-
-    def test_history_empty(self, client, s3_bucket) -> None:
-        resp = client.get("/api/v1/backup/history", headers=auth_header("power"))
-        assert resp.status_code == 200
-        assert resp.json() == []
-
-    def test_history_returns_entries(self, client, s3_bucket) -> None:
-        _upload(client, tier="power", timestamp=1000)
-        _upload(client, tier="power", timestamp=2000)
-        history = client.get(
-            "/api/v1/backup/history", headers=auth_header("power")
-        ).json()
-        assert len(history) == 2
-        # Ordered by timestamp descending
-        assert history[0]["timestamp"] == 2000
-        assert history[1]["timestamp"] == 1000
-
-    def test_history_isolated_per_user(self, client, s3_bucket) -> None:
-        """One user's backups should not appear in another user's history."""
-        _upload(client, tier="power")
-        resp = client.get("/api/v1/backup/history", headers=auth_header("team"))
-        assert resp.json() == []
-
-
-# ── TestDeleteBackup ──────────────────────────────────────────────────
-
-
-class TestDeleteBackup:
-    """DELETE /api/v1/backup/{backup_id}"""
-
-    def _get_backup_id(self, client, tier="power") -> str:
-        """Upload a backup and return its DB id from history."""
-        _upload(client, tier=tier)
-        client.get(
-            "/api/v1/backup/history", headers=auth_header(tier)
-        ).json()
-        # History returns BackupMetadata schema which doesn't have `id`.
-        # We need to look it up via a different means.
-        # Since there's only 1 backup, find via history length.
-        # Actually the schema doesn't return id — let's verify via re-download.
-        # We'll use a workaround: upload, then list history to confirm it exists,
-        # then try to delete — but we need the id...
-        # Let's check if history includes an id field.
-        # The schema is: version, timestamp, checksum, chunk_count — no id.
-        # We'll need to query the DB directly or use a known ID.
-        # For testing, we'll search history then use the DB.
-        return None  # pragma: no cover — overridden below
-
-    def test_delete_success(self, client, s3_bucket, db_session) -> None:
-        _upload(client, tier="power")
-
-        # Discover the backup_id via direct DB query
-        import asyncio
-        from sqlalchemy import select
-        from app.models import BackupMetadata
-
-        async def _get_id():
-            result = await db_session.execute(
-                select(BackupMetadata.id).where(
-                    BackupMetadata.user_id == TEST_USER_IDS["power"]
-                )
-            )
-            return result.scalar_one()
-
-        backup_id = asyncio.get_event_loop().run_until_complete(_get_id())
-
-        resp = client.delete(
-            f"/api/v1/backup/{backup_id}", headers=auth_header("power")
-        )
-        assert resp.status_code == 200
-        assert resp.json() == {"ok": True}
-
-        # History should now be empty
-        history = client.get(
-            "/api/v1/backup/history", headers=auth_header("power")
-        ).json()
-        assert history == []
-
-    def test_delete_nonexistent(self, client, s3_bucket) -> None:
-        resp = client.delete(
-            "/api/v1/backup/no-such-id", headers=auth_header("power")
-        )
-        assert resp.status_code == 404
-
-    def test_delete_other_users_backup(self, client, s3_bucket, db_session) -> None:
-        """Cannot delete another user's backup (ownership check returns 404)."""
-        _upload(client, tier="power")
-
-        import asyncio
-        from sqlalchemy import select
-        from app.models import BackupMetadata
-
-        async def _get_id():
-            result = await db_session.execute(
-                select(BackupMetadata.id).where(
-                    BackupMetadata.user_id == TEST_USER_IDS["power"]
-                )
-            )
-            return result.scalar_one()
-
-        backup_id = asyncio.get_event_loop().run_until_complete(_get_id())
-
-        # team user tries to delete power user's backup → 404
-        resp = client.delete(
-            f"/api/v1/backup/{backup_id}", headers=auth_header("team")
-        )
-        assert resp.status_code == 404

tests/test_classify_file.py

@@ -1,184 +0,0 @@
-"""Unit tests for Step 1 file classification (_classify_file).
-
-These tests call the real LLM so they require OPENAI_API_KEY / LLM env vars.
-Run with: pytest tests/test_classify_file.py -v
-
-To run a quick manual check against a real file without the full UI:
-    python -m tests.test_classify_file <path/to/file.txt> [project_name...]
-"""
-
-from __future__ import annotations
-
-import asyncio
-import sys
-
-import pytest
-
-from app.core.agent_runner import _classify_file
-
-
-# ── Fixtures ──────────────────────────────────────────────────────────────
-
-PROJECTS_SAMPLE = [
-    {
-        "id": "aaaa-0001-0000-0000-000000000001",
-        "name": "ARPA Sicilia POC",
-        "status": "active",
-        "aiSummary": "Proof of concept for AI features targeting ARPA Sicilia agency.",
-    },
-    {
-        "id": "bbbb-0002-0000-0000-000000000002",
-        "name": "SNAM AI Meeting Prep",
-        "status": "active",
-        "aiSummary": "AI-assisted preparation of meeting materials for SNAM.",
-    },
-    {
-        "id": "cccc-0003-0000-0000-000000000003",
-        "name": "SFERA+ Wave 2",
-        "status": "active",
-        "aiSummary": "Second wave of the SFERA+ whitelist project.",
-    },
-]
-
-ARPA_EMAIL = """\
-to: roberto.musso@hpe.com; luca.tondin@hpecds.com
-isImportance: normal
-hasAttachment: True
----
-## Body
-Buongiorno,
-
-In riferimento alla riunione di ieri sul POC ARPA Sicilia, vi invio il riassunto
-dei deliverable concordati:
-- Preparare demo entro il 30 marzo
-- Condividere documentazione tecnica con il team ARPA
-- Fissare call di follow-up la prossima settimana
-
-Cordiali saluti
-Roberto Marchetti
-"""
-
-SNAM_EMAIL = """\
-to: roberto.musso@hpe.com
-isImportance: high
-hasAttachment: False
----
-## Body
-Ciao,
-ti invio l'agenda per la riunione SNAM di domani.
-Per favore conferma la tua presenza.
-"""
-
-UNRELATED_EMAIL = """\
-to: roberto.musso@hpe.com
-isImportance: normal
----
-## Body
-Benvenuto nel programma HPE Employee Learning Series.
-Completa la formazione richiesta entro la fine del trimestre.
-"""
-
-
-# ── Tests ─────────────────────────────────────────────────────────────────
-
-
-@pytest.mark.asyncio
-async def test_classify_arpa_matches_existing():
-    project_id, domains, new_name = await _classify_file(
-        file_path="arpa_email.txt",
-        file_content=ARPA_EMAIL,
-        projects=PROJECTS_SAMPLE,
-        config_data_types=["tasks", "notes", "timelines"],
-    )
-    assert project_id == "aaaa-0001-0000-0000-000000000001", (
-        f"Expected ARPA project, got project_id={project_id!r} new_name={new_name!r}"
-    )
-    assert new_name is None
-
-
-@pytest.mark.asyncio
-async def test_classify_snam_matches_existing():
-    project_id, domains, new_name = await _classify_file(
-        file_path="snam_email.txt",
-        file_content=SNAM_EMAIL,
-        projects=PROJECTS_SAMPLE,
-        config_data_types=["tasks", "notes"],
-    )
-    assert project_id == "bbbb-0002-0000-0000-000000000002", (
-        f"Expected SNAM project, got project_id={project_id!r} new_name={new_name!r}"
-    )
-
-
-@pytest.mark.asyncio
-async def test_classify_unrelated_returns_new():
-    project_id, domains, new_name = await _classify_file(
-        file_path="learning_email.txt",
-        file_content=UNRELATED_EMAIL,
-        projects=PROJECTS_SAMPLE,
-        config_data_types=["tasks", "notes"],
-    )
-    assert project_id == "new"
-    assert new_name is not None  # LLM should suggest a name
-
-
-@pytest.mark.asyncio
-async def test_classify_empty_file_returns_new():
-    project_id, domains, new_name = await _classify_file(
-        file_path="empty.txt",
-        file_content="   ",
-        projects=PROJECTS_SAMPLE,
-        config_data_types=["tasks"],
-    )
-    assert project_id == "new"
-
-
-@pytest.mark.asyncio
-async def test_classify_no_projects_returns_new():
-    project_id, domains, new_name = await _classify_file(
-        file_path="arpa_email.txt",
-        file_content=ARPA_EMAIL,
-        projects=[],
-        config_data_types=["tasks", "notes"],
-    )
-    assert project_id == "new"
-    assert new_name is not None
-
-
-# ── CLI quick-test runner ─────────────────────────────────────────────────
-
-
-async def _cli_test(file_path: str, project_names: list[str]) -> None:
-    """Run Step 1 classification against a real file from the CLI."""
-    import json
-    from pathlib import Path
-
-    content = Path(file_path).read_text(encoding="utf-8", errors="replace")
-    projects = [
-        {"id": f"test-id-{i:04d}", "name": name, "status": "active", "aiSummary": ""}
-        for i, name in enumerate(project_names)
-    ]
-
-    print(f"\nClassifying: {file_path}")
-    print(f"Projects in context: {[p['name'] for p in projects]}\n")
-
-    project_id, domains, new_name = await _classify_file(
-        file_path=file_path,
-        file_content=content,
-        projects=projects,
-        config_data_types=["tasks", "notes", "timelines"],
-    )
-
-    result = {
-        "project_id": project_id,
-        "matched_name": next((p["name"] for p in projects if p["id"] == project_id), None),
-        "new_project_name": new_name,
-        "domains": domains,
-    }
-    print(json.dumps(result, indent=2, ensure_ascii=False))
-
-
-if __name__ == "__main__":
-    if len(sys.argv) < 2:
-        print("Usage: python -m tests.test_classify_file <file_path> [project_name ...]")
-        sys.exit(1)
-    asyncio.run(_cli_test(sys.argv[1], sys.argv[2:]))

tests/test_deep_agent.py

@@ -63,7 +63,7 @@ class _FakeLLM:
 async def test_run_home_uses_mocked_tool_result():
     fake_llm = _FakeLLM()
 
-    with patch("app.core.deep_agent.get_llm", return_value=fake_llm), patch(
+    with patch("app.core.deep_agent.get_agent_llm", return_value=fake_llm), patch(
         "app.core.deep_agent._all_tools", return_value=[_FakeTool()]
     ):
         out = await run_home("user-1", "list my tasks", {})
@@ -76,7 +76,7 @@ async def test_run_home_uses_mocked_tool_result():
 async def test_run_floating_stream_emits_domain_then_tokens_with_mocked_tool_result():
     fake_llm = _FakeLLM()
 
-    with patch("app.core.deep_agent.get_llm", return_value=fake_llm), patch(
+    with patch("app.core.deep_agent.get_agent_llm", return_value=fake_llm), patch(
         "app.core.deep_agent._all_tools", return_value=[_FakeTool()]
     ):
         events = []
@@ -103,7 +103,7 @@ async def test_infer_floating_domain_prefers_message_intent_over_scope_type():
                 content='{"type":"project","id":"213213-312321-312312-421321","section":"task"}'
             )
 
-    with patch("app.core.deep_agent.get_llm", return_value=_ClassifierOnlyLLM()):
+    with patch("app.core.deep_agent.get_agent_llm", return_value=_ClassifierOnlyLLM()):
         domain = await _infer_floating_domain(
             "Quali sono i miei task per il progetto X",
             {
@@ -165,7 +165,7 @@ async def test_run_floating_strips_xml_like_tags_from_final_text():
             "Mail barra in prod <task>[180faff3-507d-4d88-aba8-66f204eb59ef]</task>"
         )
 
-    with patch("app.core.deep_agent.get_llm", return_value=fake_llm), patch(
+    with patch("app.core.deep_agent.get_agent_llm", return_value=fake_llm), patch(
         "app.core.deep_agent._run_single_agent", side_effect=_fake_run_single_agent
     ):
         text, _domain = await run_floating(
@@ -187,7 +187,7 @@ async def test_run_floating_stream_strips_xml_like_tags_from_streamed_text():
         yield "token", "Hai 1 task:\\n"
         yield "token", "Mail barra in prod <task>[180faff3-507d-4d88-aba8-66f204eb59ef]</task>"
 
-    with patch("app.core.deep_agent.get_llm", return_value=fake_llm), patch(
+    with patch("app.core.deep_agent.get_agent_llm", return_value=fake_llm), patch(
         "app.core.deep_agent._run_single_agent_stream", side_effect=_fake_stream
     ):
         events = []
@@ -233,7 +233,7 @@ async def test_run_floating_stream_falls_back_to_final_response_content_when_ast
             if False:
                 yield None
 
-    with patch("app.core.deep_agent.get_llm", return_value=_NoChunkLLM()), patch(
+    with patch("app.core.deep_agent.get_agent_llm", return_value=_NoChunkLLM()), patch(
         "app.core.deep_agent._all_tools", return_value=[_FakeTool()]
     ):
         events = []
@@ -255,7 +255,7 @@ async def test_run_floating_returns_fallback_when_sanitization_would_empty_text(
     async def _fake_run_single_agent(**_kwargs):
         return "<task>[180faff3-507d-4d88-aba8-66f204eb59ef]</task>"
 
-    with patch("app.core.deep_agent.get_llm", return_value=fake_llm), patch(
+    with patch("app.core.deep_agent.get_agent_llm", return_value=fake_llm), patch(
         "app.core.deep_agent._run_single_agent", side_effect=_fake_run_single_agent
     ):
         text, _domain = await run_floating(
@@ -274,7 +274,7 @@ async def test_run_floating_stream_returns_fallback_when_sanitization_would_empt
     async def _fake_stream(**_kwargs):
         yield "token", "<task>[180faff3-507d-4d88-aba8-66f204eb59ef]</task>"
 
-    with patch("app.core.deep_agent.get_llm", return_value=fake_llm), patch(
+    with patch("app.core.deep_agent.get_agent_llm", return_value=fake_llm), patch(
         "app.core.deep_agent._run_single_agent_stream", side_effect=_fake_stream
     ):
         events = []

tests/test_device_ws.py

@@ -18,13 +18,12 @@ from datetime import datetime, timezone
 from unittest.mock import AsyncMock, MagicMock, patch
 
 import pytest
-import pytest_asyncio
 
-from app.core.device_manager import DeviceConnection, DeviceConnectionManager
+from app.core.device_manager import DeviceConnectionManager
 from app.db import get_session
 from app.main import app
 from app.models import AgentRunLog
-from tests.conftest import TEST_USER_IDS, auth_header, make_jwt
+from tests.conftest import TEST_USER_IDS, make_jwt
 
 # ---------------------------------------------------------------------------
 # Helpers
@@ -157,40 +156,6 @@ async def test_manager_unregister_cancels_pending_calls(manager, mock_ws):
     assert fut.cancelled()
 
 
-@pytest.mark.asyncio
-async def test_manager_agent_data_queue(manager, mock_ws):
-    manager.register("user1", "dev-A", mock_ws)
-    q = manager.get_agent_data_queue("user1", "run-xyz")
-    # Put a frame and get it back.
-    frame = {"type": "agent_data", "run_id": "run-xyz", "files": []}
-    await q.put(frame)
-    assert await q.get() == frame
-
-
-@pytest.mark.asyncio
-async def test_manager_agent_data_queue_creates_once(manager, mock_ws):
-    manager.register("user1", "dev-A", mock_ws)
-    q1 = manager.get_agent_data_queue("user1", "run-1")
-    q2 = manager.get_agent_data_queue("user1", "run-1")
-    assert q1 is q2
-
-
-@pytest.mark.asyncio
-async def test_manager_agent_data_queue_raises_when_offline(manager):
-    with pytest.raises(RuntimeError, match="not connected"):
-        manager.get_agent_data_queue("ghost", "run-1")
-
-
-@pytest.mark.asyncio
-async def test_manager_cleanup_agent_data_queue(manager, mock_ws):
-    manager.register("user1", "dev-A", mock_ws)
-    manager.get_agent_data_queue("user1", "run-1")
-    manager.cleanup_agent_data_queue("user1", "run-1")
-    # After cleanup a new queue is created (not the same object).
-    q_new = manager.get_agent_data_queue("user1", "run-1")
-    assert q_new is not None
-
-
 # ---------------------------------------------------------------------------
 # Integration tests — /api/v1/ws/device endpoint
 # ---------------------------------------------------------------------------
@@ -267,43 +232,6 @@ def test_ws_device_tool_result_dispatched(client):
     assert any(c["call_id"] == "call-123" for c in captured)
 
 
-def test_ws_device_agent_data_enqueued(client):
-    """agent_data frame is placed in the per-run queue by the message loop."""
-    from app.core.device_manager import device_manager as dm
-
-    token = make_jwt(tier="free")
-    user_id = TEST_USER_IDS["free"]
-
-    # Capture the queue object the message loop accesses.
-    captured_queue: list[asyncio.Queue] = []
-    original_get_queue = dm.get_agent_data_queue
-
-    def _spy_get_queue(uid, run_id):
-        q = original_get_queue(uid, run_id)
-        if not captured_queue:
-            captured_queue.append(q)
-        return q
-
-    with patch.object(dm, "get_agent_data_queue", side_effect=_spy_get_queue):
-        with patch("app.api.routes.device_ws._HEARTBEAT_INTERVAL", 9999):
-            with client.websocket_connect(f"/api/v1/ws/device?token={token}") as ws:
-                ws.send_text(_device_hello("dev-001"))
-                ws.send_text(
-                    json.dumps(
-                        {
-                            "type": "agent_data",
-                            "run_id": "run-XYZ",
-                            "files": [{"path": "/tmp/file.txt", "content": "hello"}],
-                        }
-                    )
-                )
-                ws.close()
-
-    # The queue should have received exactly one frame.
-    assert captured_queue, "queue was never accessed"
-    assert not captured_queue[0].empty()
-
-
 def test_ws_device_disconnect_marks_run_logs_as_error(client, db_session):
     """On disconnect, _mark_runs_disconnected is called with the correct user_id."""
     from app.api.routes import device_ws as _dws

tests/test_integrations.py

@@ -40,11 +40,9 @@ Coverage:
 
 from __future__ import annotations
 
-import asyncio
 import json
-import uuid
 from datetime import datetime, timezone
-from unittest.mock import AsyncMock, MagicMock, Mock, PropertyMock, patch
+from unittest.mock import AsyncMock, MagicMock, PropertyMock, patch
 
 import pytest
 

tests/test_journey_v2.py

@@ -0,0 +1,299 @@
+"""Tests for Local Agent V2 journey setup (Step 4).
+
+Covers the chatbot journey that produces a structured AgentConfig JSON
+instead of a freeform prompt_template string.
+
+Unit tests (no LLM)
+--------------------
+  4.6a  _extract_agent_config: valid JSON → returns serialised config
+  4.6b  _extract_agent_config: invalid JSON → returns None
+  4.6c  _extract_agent_config: markers absent → returns None
+  4.6d  _extract_agent_config: only START marker → returns None
+  4.6e  Session not found → done=True, agent_config=None
+  4.6f  Nudge uses AGENT_CONFIG_START/END markers (not old PROMPT_TEMPLATE)
+
+Eval test (real LLM + Langfuse scoring)
+----------------------------------------
+  4.1   Journey start explores directory → first reply contains a question
+
+Cases 4.2–4.5 (multi-turn conversations producing a full AgentConfig) are
+non-deterministic and tested manually — results tracked in Langfuse.
+
+Run:
+    pytest tests/test_journey_v2.py -v
+    pytest tests/test_journey_v2.py -v -k "4_6"          # unit only
+    pytest tests/test_journey_v2.py -v -k "eval"          # single LLM eval
+    pytest tests/test_journey_v2.py -v --journey-dir /p   # custom fixtures
+"""
+
+from __future__ import annotations
+
+import uuid
+from contextlib import nullcontext
+from pathlib import Path
+from typing import Any
+from unittest.mock import patch
+
+import pytest
+import yaml
+
+from app.api.routes.agent_setup import (
+    _CONFIG_END,
+    _CONFIG_START,
+    _MAX_TURNS,
+    _extract_agent_config,
+    _sessions,
+    handle_journey_message,
+    handle_journey_start,
+)
+from app.core.langfuse_client import get_langfuse
+from app.core.ws_context import clear_client_executor, set_client_executor
+from app.schemas import AgentConfig
+from tests.conftest import TEST_USER_IDS
+
+# ── Constants ─────────────────────────────────────────────────────────────
+
+_USER_ID = TEST_USER_IDS["power"]
+
+_DEFAULT_FIXTURE_DIR = Path(__file__).parent / "fixtures" / "journey_v2"
+
+# ── Fixture loading ───────────────────────────────────────────────────────
+
+
+def _fixtures_dir(config) -> Path:
+    override = config.getoption("--journey-dir")
+    return Path(override) if override else _DEFAULT_FIXTURE_DIR
+
+
+def _load_cases(config) -> list[dict]:
+    return yaml.safe_load(
+        (_fixtures_dir(config) / "cases.yaml").read_text(encoding="utf-8")
+    )
+
+
+def _read_data_file(filename: str, fixtures_dir: Path) -> str:
+    return (fixtures_dir / "data" / filename).read_text(encoding="utf-8")
+
+
+# ── pytest_generate_tests ─────────────────────────────────────────────────
+
+
+def pytest_generate_tests(metafunc):
+    if "journey_case" not in metafunc.fixturenames:
+        return
+    cases = _load_cases(metafunc.config)
+    metafunc.parametrize("journey_case", cases, ids=[c["id"] for c in cases])
+
+
+# ── Executor builder ──────────────────────────────────────────────────────
+
+
+def _make_fs_executor(directory_files: list[dict], fixtures_dir: Path):
+    """Return an async callback that simulates filesystem tool responses.
+
+    Matches the signature expected by ``set_client_executor`` / ``execute_on_client``:
+    receives the full ``payload`` dict and returns a result dict.
+
+    ``directory_files`` is a list of ``{path, content_file}`` dicts;
+    ``content_file`` is relative to ``fixtures_dir/data/``.
+    """
+    file_map: dict[str, str] = {
+        entry["path"]: _read_data_file(entry["content_file"], fixtures_dir)
+        for entry in directory_files
+    }
+
+    async def _executor(payload: dict) -> dict:
+        action = payload.get("action", "")
+        data = payload.get("data") or {}
+
+        if action == "list_directory":
+            return {"entries": [
+                {"type": "file", "name": p.split("/")[-1], "path": p}
+                for p in file_map
+            ]}
+
+        if action == "read_file_content":
+            path = data.get("path", "")
+            return {"content": file_map.get(path, "")}
+
+        if action == "get_file_metadata":
+            path = data.get("path", "")
+            name = path.split("/")[-1]
+            ext = "." + name.rsplit(".", 1)[-1] if "." in name else ""
+            return {"name": name, "extension": ext, "size": 1024,
+                    "createdAt": None, "modifiedAt": None}
+
+        return {}
+
+    return _executor
+
+
+# ── Journey runner helper ─────────────────────────────────────────────────
+
+
+async def _run_journey(user_id: str, case: dict, executor) -> dict[str, Any]:
+    """Drive start + all user_messages for a case. Returns the final reply dict.
+
+    Mirrors ``device_ws._handle_journey_start/message``: sets the client
+    executor (so filesystem tools work) before each handler call.
+    """
+    session_id = str(uuid.uuid4())
+    try:
+        set_client_executor(executor)
+        reply = await handle_journey_start(user_id, {
+            "agent_type": "local",
+            "directory": case["directory"],
+            "data_types": case["data_types"],
+            "session_id": session_id,
+        })
+
+        for msg in case.get("user_messages", []):
+            if reply.get("done"):
+                break
+            set_client_executor(executor)
+            reply = await handle_journey_message(user_id, {
+                "session_id": reply["session_id"],
+                "message": msg,
+            })
+    finally:
+        clear_client_executor()
+        _sessions.pop(session_id, None)
+
+    return reply
+
+
+# ── Assertion helper ──────────────────────────────────────────────────────
+
+
+def _evaluate_case(case: dict, reply: dict) -> tuple[float, str]:
+    """Return (score, comment) for a journey case given the final reply dict."""
+    if case.get("expect_question"):
+        has_q = "?" in reply.get("message", "")
+        return (1.0 if has_q else 0.0), f"first_reply_has_question={has_q}"
+
+    return 1.0, "no specific assertion"
+
+
+# ── Unit tests ────────────────────────────────────────────────────────────
+
+
+def test_4_6a_extract_valid_json():
+    """_extract_agent_config: valid JSON between markers → returns serialised config."""
+    config = AgentConfig(
+        content_types=[],
+        global_rules=["No project = no entity"],
+        data_types=["tasks"],
+    )
+    text = f"Some preamble\n{_CONFIG_START}\n{config.model_dump_json()}\n{_CONFIG_END}\nTrailing"
+    result = _extract_agent_config(text)
+    assert result is not None
+    parsed = AgentConfig.model_validate_json(result)
+    assert parsed.global_rules == ["No project = no entity"]
+
+
+def test_4_6b_extract_invalid_json():
+    """_extract_agent_config: malformed JSON between markers → returns None."""
+    text = f"{_CONFIG_START}\n{{not: valid json\n{_CONFIG_END}"
+    assert _extract_agent_config(text) is None
+
+
+def test_4_6c_extract_markers_absent():
+    """_extract_agent_config: no markers at all → returns None."""
+    assert _extract_agent_config("No markers here at all") is None
+
+
+def test_4_6d_extract_only_start_marker():
+    """_extract_agent_config: START without END → returns None."""
+    assert _extract_agent_config(f"text {_CONFIG_START} no end marker") is None
+
+
+@pytest.mark.asyncio
+async def test_4_6e_session_not_found():
+    """4.6e Session not found → done=True, agent_config=None, informative message."""
+    reply = await handle_journey_message(_USER_ID, {
+        "session_id": "nonexistent-session-id",
+        "message": "Hello",
+    })
+    assert reply["done"] is True
+    assert reply["agent_config"] is None
+    assert "not found" in reply["message"].lower() or "expired" in reply["message"].lower()
+
+
+@pytest.mark.asyncio
+async def test_4_6f_nudge_uses_new_markers():
+    """4.6f Nudge injected after max turns uses AGENT_CONFIG markers, not PROMPT_TEMPLATE."""
+    session_id = str(uuid.uuid4())
+    captured_histories: list[list[dict]] = []
+
+    async def _mock_llm(system_prompt, history, tools, **kwargs) -> str:
+        captured_histories.append(list(history))
+        # Return plain text — no markers — to trigger the nudge path.
+        return "I still need more information from you."
+
+    from app.api.routes.agent_setup import JourneySession
+
+    fake_session = JourneySession(
+        session_id=session_id,
+        user_id=_USER_ID,
+        agent_type="local",
+        directory="/test",
+        data_types=["tasks"],
+        system_prompt="system",
+        langfuse_prompt=None,
+    )
+    # Fill history to the turn limit so the next message triggers the nudge.
+    for i in range(_MAX_TURNS):
+        fake_session.history.append({"role": "user", "content": f"msg {i}"})
+        fake_session.history.append({"role": "assistant", "content": "ok"})
+    _sessions[session_id] = fake_session
+
+    try:
+        with patch("app.api.routes.agent_setup._call_llm_with_tools", side_effect=_mock_llm):
+            await handle_journey_message(_USER_ID, {
+                "session_id": session_id,
+                "message": "one more message to trigger nudge",
+            })
+    finally:
+        _sessions.pop(session_id, None)
+
+    # Second LLM call receives the nudge appended to history.
+    assert len(captured_histories) >= 2, "Expected ≥ 2 LLM calls (main reply + nudge)"
+    nudge_history = captured_histories[1]
+    user_msgs = " ".join(t["content"] for t in nudge_history if t["role"] == "user")
+    assert _CONFIG_START in user_msgs, f"Nudge must reference {_CONFIG_START}"
+    assert _CONFIG_END in user_msgs, f"Nudge must reference {_CONFIG_END}"
+    assert "PROMPT_TEMPLATE" not in user_msgs, "Old PROMPT_TEMPLATE markers must not appear in nudge"
+
+
+# ── Eval tests (real LLM + Langfuse) ─────────────────────────────────────
+
+
+@pytest.mark.asyncio
+@pytest.mark.eval
+async def test_eval_journey(journey_case, pytestconfig):
+    """Parametrized eval test — one invocation per YAML case."""
+    case: dict = journey_case
+    fixtures_dir = _fixtures_dir(pytestconfig)
+    executor = _make_fs_executor(case.get("directory_files", []), fixtures_dir)
+
+    lf = get_langfuse()
+    obs_ctx = lf.start_as_current_observation(
+        name=f"eval-journey-{case['id']}-{case.get('score_name', 'unknown').replace('.', '-')}",
+        metadata={"step": "4", "case_id": case["id"]},
+    ) if lf else nullcontext()
+
+    with obs_ctx as obs:
+        reply = await _run_journey(_USER_ID, case, executor)
+        score, comment = _evaluate_case(case, reply)
+
+        if obs is not None:
+            obs.score(
+                name=case.get("score_name", f"journey.case_{case['id']}"),
+                value=score,
+                comment=comment,
+            )
+
+    if lf:
+        lf.flush()
+
+    assert score == 1.0, f"[{case['id']}] {case.get('description', '')} — {comment}"

tests/test_memory_audit.py

@@ -0,0 +1,405 @@
+"""Tests for Phase 7 — weekly audit_memory job.
+
+Coverage:
+  1. audit_memory never raises even if inner work fails.
+  2. _scan_associative_contradictions skips when < 2 decryptable facts.
+  3. _scan_associative_contradictions calls LLM and deletes flagged rows.
+  4. _scan_associative_contradictions is a no-op when LLM fails.
+  5. _scan_associative_contradictions is a no-op when LLM returns non-list.
+  6. _canonicalize_relation_labels skips when no relation rows.
+  7. _canonicalize_relation_labels rewrites variant labels to canonical form.
+  8. _canonicalize_relation_labels is a no-op when LLM fails.
+  9. _canonicalize_relation_labels is a no-op when remap is empty.
+  10. Both helpers work correctly when Langfuse is unavailable (lf=None).
+  11. get_prompt_or_fallback called with correct Langfuse prompt names.
+"""
+
+from __future__ import annotations
+
+import json
+import uuid
+from contextlib import contextmanager, ExitStack
+from datetime import datetime, timezone
+from unittest.mock import AsyncMock, MagicMock, patch
+
+import pytest
+import pytest_asyncio
+from cryptography.fernet import Fernet
+from sqlalchemy import select
+
+from app.core.memory_maintenance import (
+    _canonicalize_relation_labels,
+    _scan_associative_contradictions,
+    audit_memory,
+)
+from app.db import get_session
+from app.main import app
+from app.models import MemoryAssociative, MemoryRelation, User
+from tests.conftest import TEST_USER_IDS
+
+PRO_USER_ID = TEST_USER_IDS["pro"]
+_FERNET_KEY = Fernet.generate_key().decode()
+_FERNET = Fernet(_FERNET_KEY.encode())
+
+
+# ── DB override ───────────────────────────────────────────────────────────────
+
+@pytest.fixture(autouse=True)
+def _override_db(db_session):
+    async def _gen():
+        yield db_session
+
+    app.dependency_overrides[get_session] = _gen
+    yield
+    app.dependency_overrides.pop(get_session, None)
+
+
+# ── Helpers ───────────────────────────────────────────────────────────────────
+
+@pytest_asyncio.fixture
+async def pro_user(db_session):
+    result = await db_session.execute(select(User).where(User.id == PRO_USER_ID))
+    user = result.scalar_one()
+    user.encryption_key = _FERNET_KEY
+    await db_session.commit()
+    return user
+
+
+def _enc(text: str) -> str:
+    return _FERNET.encrypt(text.encode()).decode()
+
+
+def _assoc_row(user_id: str, text: str) -> MemoryAssociative:
+    return MemoryAssociative(
+        id=str(uuid.uuid4()),
+        user_id=user_id,
+        content_encrypted=_enc(text),
+        updated_at=datetime.now(timezone.utc),
+    )
+
+
+def _relation_row(user_id: str, subject: str, predicate: str, obj: str) -> MemoryRelation:
+    return MemoryRelation(
+        id=str(uuid.uuid4()),
+        user_id=user_id,
+        subject_label=subject,
+        subject_type="person",
+        predicate=predicate,
+        object_label=obj,
+        object_type="company",
+        confidence=0.8,
+    )
+
+
+def _llm_response(content: str) -> MagicMock:
+    msg = MagicMock()
+    msg.content = content
+    msg.usage_metadata = {"input_tokens": 10, "output_tokens": 5, "total_tokens": 15}
+    return msg
+
+
+def _mock_llm(content: str) -> MagicMock:
+    llm = MagicMock()
+    llm.ainvoke = AsyncMock(return_value=_llm_response(content))
+    return llm
+
+
+@contextmanager
+def _patch_audit(llm_mock, lf=None, prompt_text: str = "fallback {facts}"):
+    """Context manager that patches all external deps for audit helpers."""
+    with ExitStack() as stack:
+        stack.enter_context(
+            patch("app.core.llm.get_agent_llm", return_value=llm_mock)
+        )
+        stack.enter_context(
+            patch("app.core.llm.model_for_agent", return_value="memory-auditor")
+        )
+        stack.enter_context(
+            patch("app.core.memory_maintenance.get_langfuse", return_value=lf)
+        )
+        stack.enter_context(
+            patch(
+                "app.core.memory_maintenance.get_prompt_or_fallback",
+                return_value=(prompt_text, None),
+            )
+        )
+        stack.enter_context(
+            patch(
+                "app.core.memory_maintenance.compile_prompt",
+                side_effect=lambda tmpl, obj, **kw: tmpl.format(**kw) if "{" in tmpl else tmpl,
+            )
+        )
+        yield
+
+
+# ── Test 1: audit_memory never raises ────────────────────────────────────────
+
+@pytest.mark.asyncio
+async def test_audit_memory_never_raises_on_missing_user(db_session):
+    """audit_memory with a non-existent user_id must not raise."""
+    await audit_memory(db_session, str(uuid.uuid4()))
+
+
+@pytest.mark.asyncio
+async def test_audit_memory_never_raises_on_llm_failure(db_session, pro_user):
+    """audit_memory must swallow inner exceptions."""
+    llm = MagicMock()
+    llm.ainvoke = AsyncMock(side_effect=RuntimeError("LLM down"))
+
+    with (
+        patch("app.core.llm.get_agent_llm", return_value=llm),
+        patch("app.core.llm.model_for_agent", return_value="memory-auditor"),
+        patch("app.core.memory_maintenance.get_langfuse", return_value=None),
+        patch(
+            "app.core.memory_maintenance.get_prompt_or_fallback",
+            return_value=("p {facts}", None),
+        ),
+        patch("app.core.memory_maintenance.compile_prompt", return_value="compiled"),
+    ):
+        await audit_memory(db_session, PRO_USER_ID)
+
+
+# ── Test 2: _scan skips when < 2 facts ───────────────────────────────────────
+
+@pytest.mark.asyncio
+async def test_scan_contradictions_skips_with_one_fact(db_session, pro_user):
+    row = _assoc_row(PRO_USER_ID, "Prefers morning meetings")
+    db_session.add(row)
+    await db_session.commit()
+
+    llm = MagicMock()
+    llm.ainvoke = AsyncMock(return_value=_llm_response("[]"))
+
+    with _patch_audit(llm):
+        await _scan_associative_contradictions(db_session, PRO_USER_ID, _FERNET)
+
+    llm.ainvoke.assert_not_called()
+
+
+# ── Test 3: _scan deletes flagged contradiction ───────────────────────────────
+
+@pytest.mark.asyncio
+async def test_scan_contradictions_deletes_flagged_row(db_session, pro_user):
+    keep = _assoc_row(PRO_USER_ID, "Prefers morning meetings")
+    drop = _assoc_row(PRO_USER_ID, "Never schedules before noon")
+    db_session.add(keep)
+    db_session.add(drop)
+    await db_session.commit()
+
+    deletion_payload = json.dumps([{"delete": drop.id, "reason": "contradicts morning pref"}])
+    llm = _mock_llm(deletion_payload)
+
+    with _patch_audit(llm, prompt_text="p {facts}"):
+        await _scan_associative_contradictions(db_session, PRO_USER_ID, _FERNET)
+
+    result = await db_session.execute(
+        select(MemoryAssociative).where(MemoryAssociative.user_id == PRO_USER_ID)
+    )
+    remaining = result.scalars().all()
+    remaining_ids = {r.id for r in remaining}
+    assert keep.id in remaining_ids
+    assert drop.id not in remaining_ids
+
+
+# ── Test 4: _scan is no-op on LLM failure ────────────────────────────────────
+
+@pytest.mark.asyncio
+async def test_scan_contradictions_noop_on_llm_failure(db_session, pro_user):
+    for text in ("Fact A", "Fact B"):
+        db_session.add(_assoc_row(PRO_USER_ID, text))
+    await db_session.commit()
+
+    llm = MagicMock()
+    llm.ainvoke = AsyncMock(side_effect=RuntimeError("LLM down"))
+
+    with _patch_audit(llm, prompt_text="p {facts}"):
+        await _scan_associative_contradictions(db_session, PRO_USER_ID, _FERNET)
+
+    result = await db_session.execute(
+        select(MemoryAssociative).where(MemoryAssociative.user_id == PRO_USER_ID)
+    )
+    assert len(result.scalars().all()) == 2
+
+
+# ── Test 5: _scan is no-op when LLM returns non-list ─────────────────────────
+
+@pytest.mark.asyncio
+async def test_scan_contradictions_noop_on_non_list_response(db_session, pro_user):
+    for text in ("Fact A", "Fact B"):
+        db_session.add(_assoc_row(PRO_USER_ID, text))
+    await db_session.commit()
+
+    llm = _mock_llm('"unexpected string"')
+
+    with _patch_audit(llm, prompt_text="p {facts}"):
+        await _scan_associative_contradictions(db_session, PRO_USER_ID, _FERNET)
+
+    result = await db_session.execute(
+        select(MemoryAssociative).where(MemoryAssociative.user_id == PRO_USER_ID)
+    )
+    assert len(result.scalars().all()) == 2
+
+
+# ── Test 6: _canonicalize skips when no relations ────────────────────────────
+
+@pytest.mark.asyncio
+async def test_canonicalize_skips_when_no_relations(db_session, pro_user):
+    llm = MagicMock()
+    llm.ainvoke = AsyncMock(return_value=_llm_response("[]"))
+
+    with _patch_audit(llm, prompt_text="p {labels}"):
+        await _canonicalize_relation_labels(db_session, PRO_USER_ID)
+
+    llm.ainvoke.assert_not_called()
+
+
+# ── Test 7: _canonicalize rewrites variant labels ────────────────────────────
+
+@pytest.mark.asyncio
+async def test_canonicalize_rewrites_variant_labels(db_session, pro_user):
+    row_a = _relation_row(PRO_USER_ID, "giulia", "works_at", "Acme")
+    row_b = _relation_row(PRO_USER_ID, "Giulia R.", "reports_to", "Marco")
+    row_c = _relation_row(PRO_USER_ID, "Marco", "manages", "Giulia")
+    db_session.add(row_a)
+    db_session.add(row_b)
+    db_session.add(row_c)
+    await db_session.commit()
+
+    groups = json.dumps([
+        {"canonical": "Giulia", "variants": ["giulia", "Giulia R."]}
+    ])
+    llm = _mock_llm(groups)
+
+    with _patch_audit(llm, prompt_text="p {labels}"):
+        await _canonicalize_relation_labels(db_session, PRO_USER_ID)
+
+    await db_session.refresh(row_a)
+    await db_session.refresh(row_b)
+    await db_session.refresh(row_c)
+
+    assert row_a.subject_label == "Giulia"
+    assert row_b.subject_label == "Giulia"
+    assert row_c.object_label == "Giulia"
+    assert row_c.subject_label == "Marco"
+
+
+# ── Test 8: _canonicalize is no-op on LLM failure ────────────────────────────
+
+@pytest.mark.asyncio
+async def test_canonicalize_noop_on_llm_failure(db_session, pro_user):
+    row = _relation_row(PRO_USER_ID, "giulia", "works_at", "Acme")
+    db_session.add(row)
+    await db_session.commit()
+
+    llm = MagicMock()
+    llm.ainvoke = AsyncMock(side_effect=RuntimeError("LLM down"))
+
+    with _patch_audit(llm, prompt_text="p {labels}"):
+        await _canonicalize_relation_labels(db_session, PRO_USER_ID)
+
+    await db_session.refresh(row)
+    assert row.subject_label == "giulia"
+
+
+# ── Test 9: _canonicalize is no-op when remap is empty ───────────────────────
+
+@pytest.mark.asyncio
+async def test_canonicalize_noop_when_remap_empty(db_session, pro_user):
+    row = _relation_row(PRO_USER_ID, "Giulia", "works_at", "Acme")
+    db_session.add(row)
+    await db_session.commit()
+
+    llm = _mock_llm("[]")
+
+    with _patch_audit(llm, prompt_text="p {labels}"):
+        await _canonicalize_relation_labels(db_session, PRO_USER_ID)
+
+    await db_session.refresh(row)
+    assert row.subject_label == "Giulia"
+
+
+# ── Test 10: both helpers work without Langfuse ───────────────────────────────
+
+@pytest.mark.asyncio
+async def test_scan_works_without_langfuse(db_session, pro_user):
+    keep = _assoc_row(PRO_USER_ID, "Prefers dark mode")
+    drop = _assoc_row(PRO_USER_ID, "Prefers light mode")
+    db_session.add(keep)
+    db_session.add(drop)
+    await db_session.commit()
+
+    deletion_payload = json.dumps([{"delete": drop.id, "reason": "contradicts dark mode"}])
+    llm = _mock_llm(deletion_payload)
+
+    with _patch_audit(llm, lf=None, prompt_text="p {facts}"):
+        await _scan_associative_contradictions(db_session, PRO_USER_ID, _FERNET)
+
+    result = await db_session.execute(
+        select(MemoryAssociative).where(MemoryAssociative.user_id == PRO_USER_ID)
+    )
+    remaining_ids = {r.id for r in result.scalars().all()}
+    assert keep.id in remaining_ids
+    assert drop.id not in remaining_ids
+
+
+@pytest.mark.asyncio
+async def test_canonicalize_works_without_langfuse(db_session, pro_user):
+    row = _relation_row(PRO_USER_ID, "giulia", "works_at", "Acme")
+    db_session.add(row)
+    db_session.add(_relation_row(PRO_USER_ID, "Marco", "manages", "Giulia"))
+    await db_session.commit()
+
+    groups = json.dumps([{"canonical": "Giulia", "variants": ["giulia"]}])
+    llm = _mock_llm(groups)
+
+    with _patch_audit(llm, lf=None, prompt_text="p {labels}"):
+        await _canonicalize_relation_labels(db_session, PRO_USER_ID)
+
+    await db_session.refresh(row)
+    assert row.subject_label == "Giulia"
+
+
+# ── Test 11: correct Langfuse prompt names used ───────────────────────────────
+
+@pytest.mark.asyncio
+async def test_scan_uses_correct_langfuse_prompt_name(db_session, pro_user):
+    for text in ("Fact A", "Fact B"):
+        db_session.add(_assoc_row(PRO_USER_ID, text))
+    await db_session.commit()
+
+    llm = _mock_llm("[]")
+    mock_get_prompt = MagicMock(return_value=("p {facts}", None))
+
+    with (
+        patch("app.core.llm.get_agent_llm", return_value=llm),
+        patch("app.core.llm.model_for_agent", return_value="memory-auditor"),
+        patch("app.core.memory_maintenance.get_langfuse", return_value=None),
+        patch("app.core.memory_maintenance.get_prompt_or_fallback", mock_get_prompt),
+        patch("app.core.memory_maintenance.compile_prompt", return_value="compiled"),
+    ):
+        await _scan_associative_contradictions(db_session, PRO_USER_ID, _FERNET)
+
+    mock_get_prompt.assert_called_once()
+    assert mock_get_prompt.call_args[0][0] == "memory_audit_contradictions"
+
+
+@pytest.mark.asyncio
+async def test_canonicalize_uses_correct_langfuse_prompt_name(db_session, pro_user):
+    db_session.add(_relation_row(PRO_USER_ID, "Giulia", "works_at", "Acme"))
+    db_session.add(_relation_row(PRO_USER_ID, "Marco", "manages", "Acme"))
+    await db_session.commit()
+
+    llm = _mock_llm("[]")
+    mock_get_prompt = MagicMock(return_value=("p {labels}", None))
+
+    with (
+        patch("app.core.llm.get_agent_llm", return_value=llm),
+        patch("app.core.llm.model_for_agent", return_value="memory-auditor"),
+        patch("app.core.memory_maintenance.get_langfuse", return_value=None),
+        patch("app.core.memory_maintenance.get_prompt_or_fallback", mock_get_prompt),
+        patch("app.core.memory_maintenance.compile_prompt", return_value="compiled"),
+    ):
+        await _canonicalize_relation_labels(db_session, PRO_USER_ID)
+
+    mock_get_prompt.assert_called_once()
+    assert mock_get_prompt.call_args[0][0] == "memory_audit_canonicalize"

tests/test_memory_extraction.py

@@ -0,0 +1,345 @@
+"""Tests for Phase 2 — Mem0-style Extract/Update pipeline.
+
+Coverage:
+  2.1 extract_candidates returns valid ExtractionResult with mocked LLM.
+  2.2 decide_action — all 4 branches (ADD/UPDATE/DELETE/NOOP + empty existing).
+  2.3 run_extraction end-to-end with mocked LLM writes expected rows.
+  2.4 _dispatch_extraction — Pro user triggers realtime task; Free enqueues row.
+"""
+
+from __future__ import annotations
+
+import json
+import uuid
+from unittest.mock import AsyncMock, MagicMock, patch
+
+import pytest
+import pytest_asyncio
+from cryptography.fernet import Fernet
+from sqlalchemy import select
+
+from app.core.memory_extraction import (
+    ExtractionResult,
+    MemoryCandidate,
+    decide_action,
+    extract_candidates,
+    run_extraction,
+)
+from app.core.memory_middleware import MemoryMiddleware
+from app.db import get_session
+from app.main import app
+from app.models import ExtractionQueue, MemoryCore, User
+from tests.conftest import TEST_USER_IDS
+
+
+PRO_USER_ID = TEST_USER_IDS["pro"]
+FREE_USER_ID = TEST_USER_IDS["free"]
+_FERNET_KEY = Fernet.generate_key().decode()
+
+
+# ── DB override ───────────────────────────────────────────────────────────────
+
+@pytest.fixture(autouse=True)
+def _override_db(db_session):
+    async def _gen():
+        yield db_session
+
+    app.dependency_overrides[get_session] = _gen
+    yield
+    app.dependency_overrides.pop(get_session, None)
+
+
+# ── Helpers ───────────────────────────────────────────────────────────────────
+
+@pytest_asyncio.fixture
+async def pro_user(db_session):
+    """Update the seeded pro user to have an encryption_key."""
+    result = await db_session.execute(select(User).where(User.id == PRO_USER_ID))
+    user = result.scalar_one()
+    user.encryption_key = _FERNET_KEY
+    await db_session.commit()
+    return user
+
+
+@pytest_asyncio.fixture
+async def free_user(db_session):
+    """Update the seeded free user to have an encryption_key."""
+    result = await db_session.execute(select(User).where(User.id == FREE_USER_ID))
+    user = result.scalar_one()
+    user.encryption_key = _FERNET_KEY
+    await db_session.commit()
+    return user
+
+
+def _make_llm_response(content: str) -> MagicMock:
+    msg = MagicMock()
+    msg.content = content
+    msg.usage_metadata = {"input_tokens": 10, "output_tokens": 5, "total_tokens": 15}
+    return msg
+
+
+# ── TASK 2.1 — extract_candidates ────────────────────────────────────────────
+
+@pytest.mark.asyncio
+async def test_extract_candidates_returns_valid_result():
+    payload = {
+        "candidates": [
+            {
+                "type": "fact",
+                "content": "User's CFO is Giulia",
+                "target_tier": "core",
+                "subject": None,
+                "predicate": None,
+                "object": None,
+                "confidence": 0.85,
+            }
+        ]
+    }
+    mock_response = _make_llm_response(json.dumps(payload))
+
+    with (
+        patch("app.core.memory_extraction.get_agent_llm") as mock_get_llm,
+        patch("app.core.memory_extraction.get_langfuse", return_value=None),
+        patch("app.core.memory_extraction.get_prompt_or_fallback") as mock_prompt,
+    ):
+        mock_prompt.return_value = (
+            "system prompt {last_turn} {core_memory} {recent_episodes}",
+            None,
+        )
+        llm_instance = MagicMock()
+        llm_instance.bind.return_value = llm_instance
+        llm_instance.ainvoke = AsyncMock(return_value=mock_response)
+        mock_get_llm.return_value = llm_instance
+
+        result = await extract_candidates(
+            last_turn="User: My CFO is Giulia\nAssistant: Noted.",
+            core_memory={},
+            recent_episodes=[],
+        )
+
+    assert isinstance(result, ExtractionResult)
+    assert len(result.candidates) == 1
+    assert result.candidates[0].type == "fact"
+    assert "Giulia" in result.candidates[0].content
+    assert result.candidates[0].confidence == 0.85
+
+
+@pytest.mark.asyncio
+async def test_extract_candidates_returns_empty_on_llm_failure():
+    with (
+        patch("app.core.memory_extraction.get_agent_llm") as mock_get_llm,
+        patch("app.core.memory_extraction.get_langfuse", return_value=None),
+        patch("app.core.memory_extraction.get_prompt_or_fallback") as mock_prompt,
+    ):
+        mock_prompt.return_value = ("prompt {last_turn} {core_memory} {recent_episodes}", None)
+        llm_instance = MagicMock()
+        llm_instance.bind.return_value = llm_instance
+        llm_instance.ainvoke = AsyncMock(side_effect=RuntimeError("LLM down"))
+        mock_get_llm.return_value = llm_instance
+
+        result = await extract_candidates("turn", {}, [])
+
+    assert isinstance(result, ExtractionResult)
+    assert result.candidates == []
+
+
+# ── TASK 2.2 — decide_action ─────────────────────────────────────────────────
+
+@pytest.mark.asyncio
+async def test_decide_action_add_when_no_existing():
+    candidate = MemoryCandidate(type="fact", content="CFO is Giulia", target_tier="core")
+    action = await decide_action(candidate, existing=[])
+    assert action == "ADD"
+
+
+@pytest.mark.asyncio
+async def test_decide_action_noop():
+    candidate = MemoryCandidate(type="fact", content="CFO is Giulia", target_tier="core")
+    mock_response = _make_llm_response("NOOP")
+
+    with (
+        patch("app.core.memory_extraction.get_agent_llm") as mock_get_llm,
+        patch("app.core.memory_extraction.get_langfuse", return_value=None),
+        patch("app.core.memory_extraction.get_prompt_or_fallback") as mock_prompt,
+    ):
+        mock_prompt.return_value = ("p {candidate} {existing_memories}", None)
+        llm_instance = MagicMock()
+        llm_instance.ainvoke = AsyncMock(return_value=mock_response)
+        mock_get_llm.return_value = llm_instance
+
+        action = await decide_action(candidate, existing=["CFO is Giulia"])
+
+    assert action == "NOOP"
+
+
+@pytest.mark.asyncio
+async def test_decide_action_update():
+    candidate = MemoryCandidate(type="fact", content="CFO is Marco", target_tier="core")
+    mock_response = _make_llm_response("UPDATE")
+
+    with (
+        patch("app.core.memory_extraction.get_agent_llm") as mock_get_llm,
+        patch("app.core.memory_extraction.get_langfuse", return_value=None),
+        patch("app.core.memory_extraction.get_prompt_or_fallback") as mock_prompt,
+    ):
+        mock_prompt.return_value = ("p {candidate} {existing_memories}", None)
+        llm_instance = MagicMock()
+        llm_instance.ainvoke = AsyncMock(return_value=mock_response)
+        mock_get_llm.return_value = llm_instance
+
+        action = await decide_action(candidate, existing=["CFO is Giulia"])
+
+    assert action == "UPDATE"
+
+
+@pytest.mark.asyncio
+async def test_decide_action_delete():
+    candidate = MemoryCandidate(type="fact", content="No longer have a CFO", target_tier="core")
+    mock_response = _make_llm_response("DELETE")
+
+    with (
+        patch("app.core.memory_extraction.get_agent_llm") as mock_get_llm,
+        patch("app.core.memory_extraction.get_langfuse", return_value=None),
+        patch("app.core.memory_extraction.get_prompt_or_fallback") as mock_prompt,
+    ):
+        mock_prompt.return_value = ("p {candidate} {existing_memories}", None)
+        llm_instance = MagicMock()
+        llm_instance.ainvoke = AsyncMock(return_value=mock_response)
+        mock_get_llm.return_value = llm_instance
+
+        action = await decide_action(candidate, existing=["CFO is Giulia"])
+
+    assert action == "DELETE"
+
+
+@pytest.mark.asyncio
+async def test_decide_action_defaults_add_on_llm_failure():
+    candidate = MemoryCandidate(type="fact", content="CFO is Marco", target_tier="core")
+
+    with (
+        patch("app.core.memory_extraction.get_agent_llm") as mock_get_llm,
+        patch("app.core.memory_extraction.get_langfuse", return_value=None),
+        patch("app.core.memory_extraction.get_prompt_or_fallback") as mock_prompt,
+    ):
+        mock_prompt.return_value = ("p {candidate} {existing_memories}", None)
+        llm_instance = MagicMock()
+        llm_instance.ainvoke = AsyncMock(side_effect=RuntimeError("LLM down"))
+        mock_get_llm.return_value = llm_instance
+
+        action = await decide_action(candidate, existing=["old memory"])
+
+    assert action == "ADD"
+
+
+# ── TASK 2.3 — run_extraction end-to-end ─────────────────────────────────────
+
+@pytest.mark.asyncio
+async def test_run_extraction_writes_core_candidate(db_session, pro_user):
+    """'My CFO is Giulia' → fact candidate → core row written."""
+    fact_payload = {
+        "candidates": [
+            {
+                "type": "fact",
+                "content": "User prefers morning meetings",
+                "target_tier": "core",
+                "confidence": 0.8,
+            }
+        ]
+    }
+
+    def _mock_llm_response(content: str):
+        msg = MagicMock()
+        msg.content = content
+        msg.usage_metadata = {}
+        return msg
+
+    call_count = 0
+
+    async def _ainvoke_side_effect(messages):
+        nonlocal call_count
+        call_count += 1
+        if call_count == 1:
+            # extract_candidates call
+            return _mock_llm_response(json.dumps(fact_payload))
+        # decide_action — no existing → short-circuits to ADD without LLM
+        return _mock_llm_response("ADD")
+
+    with (
+        patch("app.core.memory_extraction.get_agent_llm") as mock_get_llm,
+        patch("app.core.memory_extraction.get_langfuse", return_value=None),
+        patch(
+            "app.core.memory_extraction.get_prompt_or_fallback",
+            side_effect=lambda name, fb: (
+                ("p {last_turn} {core_memory} {recent_episodes}", None)
+                if name == "memory_extraction"
+                else ("p {candidate} {existing_memories}", None)
+            ),
+        ),
+    ):
+        llm_instance = MagicMock()
+        llm_instance.bind.return_value = llm_instance
+        llm_instance.ainvoke = AsyncMock(side_effect=_ainvoke_side_effect)
+        mock_get_llm.return_value = llm_instance
+
+        await run_extraction(
+            db=db_session,
+            user_id=PRO_USER_ID,
+            last_user_msg="My CFO is Giulia",
+            last_assistant_msg="Noted, I will remember that.",
+            session_id="test-session",
+        )
+
+    # core row should exist
+    result = await db_session.execute(
+        select(MemoryCore).where(MemoryCore.user_id == PRO_USER_ID)
+    )
+    rows = result.scalars().all()
+    assert len(rows) >= 1
+    fernet = Fernet(_FERNET_KEY.encode())
+    values = [fernet.decrypt(r.value_encrypted.encode()).decode() for r in rows]
+    assert any("morning meetings" in v for v in values)
+
+
+# ── TASK 2.4 — dispatch ───────────────────────────────────────────────────────
+
+@pytest.mark.asyncio
+async def test_dispatch_realtime_for_pro(db_session, pro_user):
+    """Pro user: asyncio.create_task called (not queue row)."""
+    middleware = MemoryMiddleware(db_session)
+
+    with (
+        patch("app.core.memory_middleware.asyncio.create_task") as mock_task,
+        patch("app.billing.tier_manager.tier_manager.check_feature", return_value=True),
+    ):
+        await middleware._dispatch_extraction(
+            user_id=PRO_USER_ID,
+            episode_id=str(uuid.uuid4()),
+            last_user_msg="hello",
+            last_assistant_msg="hi",
+            session_id=None,
+        )
+
+    mock_task.assert_called_once()
+
+
+@pytest.mark.asyncio
+async def test_dispatch_queue_for_free(db_session, free_user):
+    """Free user: ExtractionQueue row inserted."""
+    middleware = MemoryMiddleware(db_session)
+    ep_id = str(uuid.uuid4())
+
+    with patch("app.billing.tier_manager.tier_manager.check_feature", return_value=False):
+        await middleware._dispatch_extraction(
+            user_id=FREE_USER_ID,
+            episode_id=ep_id,
+            last_user_msg="hello",
+            last_assistant_msg="hi",
+            session_id=None,
+        )
+
+    result = await db_session.execute(
+        select(ExtractionQueue).where(ExtractionQueue.user_id == FREE_USER_ID)
+    )
+    rows = result.scalars().all()
+    assert len(rows) == 1
+    assert rows[0].episode_id == ep_id

tests/test_memory_middleware.py

@@ -12,14 +12,15 @@ from __future__ import annotations
 
 import json
 import uuid
-from unittest.mock import patch
+from unittest.mock import AsyncMock, MagicMock, patch
 
 import pytest
 import pytest_asyncio
 from cryptography.fernet import Fernet
 from sqlalchemy import select
 
-from app.core.memory_middleware import MemoryMiddleware, _PROACTIVE_CONFIDENCE_THRESHOLD
+from app.core.embeddings import embed_text
+from app.core.memory_middleware import MemoryMiddleware
 from app.db import get_session
 from app.main import app
 from app.models import (
@@ -341,3 +342,33 @@ def test_home_request_calls_memory_middleware(client):
     stored_session_id, stored_message = store_calls[0][1], store_calls[0][2]
     assert stored_session_id == session_id
     assert stored_message == "Show tasks"
+
+
+# ── embed_text ─────────────────────────────────────────────────────────────────
+
+@pytest.mark.asyncio
+async def test_embed_text_returns_1536_floats():
+    """embed_text returns a 1536-dim float list when OpenAI responds successfully."""
+    fake_embedding = [0.1] * 1536
+
+    mock_response = MagicMock()
+    mock_response.data = [MagicMock(embedding=fake_embedding)]
+
+    mock_client = MagicMock()
+    mock_client.embeddings.create = AsyncMock(return_value=mock_response)
+
+    with patch("app.core.embeddings.AsyncOpenAI", return_value=mock_client):
+        result = await embed_text("test text")
+
+    assert result is not None
+    assert len(result) == 1536
+    assert all(isinstance(x, float) for x in result)
+
+
+@pytest.mark.asyncio
+async def test_embed_text_returns_none_on_failure():
+    """embed_text returns None when OpenAI raises; must not propagate the exception."""
+    with patch("app.core.embeddings.AsyncOpenAI", side_effect=Exception("no key")):
+        result = await embed_text("test text")
+
+    assert result is None

tests/test_memory_models.py

@@ -7,10 +7,9 @@ column is stored as JSON in tests (SQLite-compatible).
 from __future__ import annotations
 
 import uuid
-from datetime import datetime, timezone
+from datetime import datetime
 
 import pytest
-import pytest_asyncio
 from cryptography.fernet import Fernet
 from sqlalchemy import select
 

tests/test_memory_proactive.py

@@ -0,0 +1,153 @@
+"""Tests for Phase 5 — proactive hints surfacing.
+
+Coverage:
+  1. _proactive_hints_injection returns correct section for seeded hints
+  2. _proactive_hints_injection returns empty string when no hints
+  3. enrich_context includes proactive_hints key from MemoryProactive row
+  4. System prompt includes proactive line when row exists + confidence >= threshold
+  5. TierManager.check_feature returns True for power/team, False for free/pro
+"""
+
+from __future__ import annotations
+
+import uuid
+
+import pytest
+import pytest_asyncio
+from cryptography.fernet import Fernet
+from sqlalchemy import select
+
+from app.billing.tier_manager import tier_manager
+from app.core.deep_agent import _proactive_hints_injection
+from app.core.memory_middleware import MemoryMiddleware
+from app.db import get_session
+from app.main import app
+from app.models import MemoryProactive, User
+from tests.conftest import TEST_USER_IDS
+
+
+USER_ID = TEST_USER_IDS["power"]
+_FERNET_KEY = Fernet.generate_key().decode()
+
+
+# ── DB override ───────────────────────────────────────────────────────────────
+
+@pytest.fixture(autouse=True)
+def _override_db(db_session):
+    async def _gen():
+        yield db_session
+
+    app.dependency_overrides[get_session] = _gen
+    yield
+    app.dependency_overrides.pop(get_session, None)
+
+
+# ── Fixtures ──────────────────────────────────────────────────────────────────
+
+@pytest_asyncio.fixture
+async def user_with_key(db_session):
+    result = await db_session.execute(select(User).where(User.id == USER_ID))
+    user = result.scalar_one()
+    user.encryption_key = _FERNET_KEY
+    await db_session.commit()
+    return user
+
+
+def _enc(plaintext: str) -> str:
+    return Fernet(_FERNET_KEY.encode()).encrypt(plaintext.encode()).decode()
+
+
+# ── _proactive_hints_injection unit tests ─────────────────────────────────────
+
+def test_proactive_hints_injection_with_hints():
+    context = {"proactive_hints": ["Works late on Thursdays", "Prefers bullet points"]}
+    result = _proactive_hints_injection(context)
+    assert "I noticed" in result
+    assert "Works late on Thursdays" in result
+    assert "Prefers bullet points" in result
+
+
+def test_proactive_hints_injection_empty():
+    assert _proactive_hints_injection({}) == ""
+    assert _proactive_hints_injection({"proactive_hints": []}) == ""
+    assert _proactive_hints_injection({"proactive_hints": None}) == ""
+
+
+def test_proactive_hints_injection_truncates_long_hints():
+    hints = ["x" * 200] * 10
+    result = _proactive_hints_injection({"proactive_hints": hints})
+    assert len(result) <= 600
+    assert result.endswith("...")
+
+
+# ── enrich_context includes proactive hints ───────────────────────────────────
+
+@pytest.mark.asyncio
+async def test_enrich_context_returns_proactive_hints(db_session, user_with_key):
+    pattern = "Always checks tasks before meetings"
+    db_session.add(MemoryProactive(
+        id=str(uuid.uuid4()),
+        user_id=USER_ID,
+        pattern_encrypted=_enc(pattern),
+        confidence=0.8,
+        source="inferred",
+    ))
+    await db_session.commit()
+
+    middleware = MemoryMiddleware(db_session)
+    ctx = await middleware.enrich_context(USER_ID, "test message")
+
+    assert "proactive_hints" in ctx
+    assert pattern in ctx["proactive_hints"]
+
+
+@pytest.mark.asyncio
+async def test_enrich_context_excludes_low_confidence_proactive(db_session, user_with_key):
+    pattern = "Low confidence pattern"
+    db_session.add(MemoryProactive(
+        id=str(uuid.uuid4()),
+        user_id=USER_ID,
+        pattern_encrypted=_enc(pattern),
+        confidence=0.1,
+        source="inferred",
+    ))
+    await db_session.commit()
+
+    middleware = MemoryMiddleware(db_session)
+    ctx = await middleware.enrich_context(USER_ID, "test message")
+
+    hints = ctx.get("proactive_hints", [])
+    assert pattern not in hints
+
+
+# ── proactive hints appear in system prompt string ───────────────────────────
+
+@pytest.mark.asyncio
+async def test_proactive_hints_in_system_prompt_string(db_session, user_with_key):
+    pattern = "Frequently requests end-of-day summaries"
+    db_session.add(MemoryProactive(
+        id=str(uuid.uuid4()),
+        user_id=USER_ID,
+        pattern_encrypted=_enc(pattern),
+        confidence=0.75,
+        source="inferred",
+    ))
+    await db_session.commit()
+
+    middleware = MemoryMiddleware(db_session)
+    ctx = await middleware.enrich_context(USER_ID, "summarize my day")
+
+    system_prompt_suffix = _proactive_hints_injection(ctx)
+    assert pattern in system_prompt_suffix
+
+
+# ── Tier gate ─────────────────────────────────────────────────────────────────
+
+@pytest.mark.parametrize("tier,expected", [
+    ("free", False),
+    ("pro", False),
+    ("power", True),
+    ("team", True),
+])
+def test_proactive_mining_tier_gate(tier, expected):
+    assert tier_manager.check_feature(tier, "proactive_mining") == expected

tests/test_memory_relations.py

@@ -0,0 +1,220 @@
+"""Tests for Phase 3 — relational tier (Mem0g-light).
+
+Coverage:
+  1. upsert_relation inserts a row and query_relations returns it
+  2. upsert_relation updates existing row on duplicate (subject/predicate/object)
+  3. tier gating: Free user gets empty list from query_relations + enrich_context
+  4. enrich_context includes relational_memory key for Pro user
+  5. decay_relations decays confidence and prunes rows below threshold
+"""
+
+from __future__ import annotations
+
+import uuid
+from datetime import datetime, timedelta, timezone
+from unittest.mock import patch
+
+import pytest
+import pytest_asyncio
+from cryptography.fernet import Fernet
+from sqlalchemy import select
+
+from app.core.memory_maintenance import decay_relations
+from app.core.memory_middleware import MemoryMiddleware
+from app.db import get_session
+from app.main import app
+from app.models import MemoryRelation, User
+from tests.conftest import TEST_USER_IDS
+
+PRO_USER_ID = TEST_USER_IDS["pro"]
+FREE_USER_ID = TEST_USER_IDS["free"]
+_FERNET_KEY = Fernet.generate_key().decode()
+
+
+# ── DB override ───────────────────────────────────────────────────────────────
+
+@pytest.fixture(autouse=True)
+def _override_db(db_session):
+    async def _gen():
+        yield db_session
+
+    app.dependency_overrides[get_session] = _gen
+    yield
+    app.dependency_overrides.pop(get_session, None)
+
+
+@pytest_asyncio.fixture
+async def pro_user_with_key(db_session):
+    """Set encryption_key on the pro test user so Fernet works."""
+    result = await db_session.execute(select(User).where(User.id == PRO_USER_ID))
+    user = result.scalar_one()
+    user.encryption_key = _FERNET_KEY
+    await db_session.commit()
+    return user
+
+
+@pytest_asyncio.fixture
+async def free_user_with_key(db_session):
+    """Set encryption_key on the free test user."""
+    result = await db_session.execute(select(User).where(User.id == FREE_USER_ID))
+    user = result.scalar_one()
+    user.encryption_key = _FERNET_KEY
+    await db_session.commit()
+    return user
+
+
+# ── Tests ─────────────────────────────────────────────────────────────────────
+
+@pytest.mark.asyncio
+async def test_upsert_relation_inserts_and_queries(db_session, pro_user_with_key):
+    """upsert_relation inserts a row; query_relations returns it."""
+    mm = MemoryMiddleware(db_session)
+    await mm.upsert_relation(
+        PRO_USER_ID,
+        subject="Giulia",
+        subject_type="person",
+        predicate="works_at",
+        object_="Acme Corp",
+        object_type="company",
+        confidence=0.9,
+    )
+    rows = await mm.query_relations(PRO_USER_ID, subject="Giulia")
+    assert len(rows) == 1
+    assert rows[0].subject_label == "Giulia"
+    assert rows[0].predicate == "works_at"
+    assert rows[0].object_label == "Acme Corp"
+    assert abs(rows[0].confidence - 0.9) < 0.001
+
+
+@pytest.mark.asyncio
+async def test_upsert_relation_updates_on_duplicate(db_session, pro_user_with_key):
+    """Second upsert on same triple updates confidence and last_confirmed_at."""
+    mm = MemoryMiddleware(db_session)
+    await mm.upsert_relation(
+        PRO_USER_ID,
+        subject="Marco",
+        subject_type="person",
+        predicate="stakeholder_of",
+        object_="Project Nexus",
+        object_type="project",
+        confidence=0.7,
+    )
+    await mm.upsert_relation(
+        PRO_USER_ID,
+        subject="Marco",
+        subject_type="person",
+        predicate="stakeholder_of",
+        object_="Project Nexus",
+        object_type="project",
+        confidence=0.95,
+    )
+    rows = await mm.query_relations(PRO_USER_ID, subject="Marco")
+    # Only one row despite two upserts
+    assert len(rows) == 1
+    assert abs(rows[0].confidence - 0.95) < 0.001
+    assert rows[0].last_confirmed_at is not None
+
+
+@pytest.mark.asyncio
+async def test_free_tier_relation_skipped(db_session, free_user_with_key):
+    """Free user: upsert_relation is silently skipped (no row created)."""
+    mm = MemoryMiddleware(db_session)
+    await mm.upsert_relation(
+        FREE_USER_ID,
+        subject="Alice",
+        subject_type="person",
+        predicate="reports_to",
+        object_="Bob",
+        object_type="person",
+        confidence=0.8,
+    )
+    rows = await mm.query_relations(FREE_USER_ID, subject="Alice")
+    assert rows == []
+
+
+@pytest.mark.asyncio
+async def test_enrich_context_includes_relational_memory(db_session, pro_user_with_key):
+    """enrich_context includes relational_memory key for Pro user."""
+    mm = MemoryMiddleware(db_session)
+    await mm.upsert_relation(
+        PRO_USER_ID,
+        subject="Elena",
+        subject_type="person",
+        predicate="cfo_of",
+        object_="StartupXYZ",
+        object_type="company",
+        confidence=0.85,
+    )
+
+    with patch("app.core.memory_middleware.MemoryMiddleware._load_associative", return_value=[]):
+        ctx = await mm.enrich_context(PRO_USER_ID, "who is Elena?")
+
+    assert "relational_memory" in ctx
+    assert any("Elena" in r for r in ctx["relational_memory"])
+
+
+@pytest.mark.asyncio
+async def test_enrich_context_relational_empty_for_free(db_session, free_user_with_key):
+    """Free user: relational_memory is empty list in enrich_context."""
+    mm = MemoryMiddleware(db_session)
+
+    with patch("app.core.memory_middleware.MemoryMiddleware._load_associative", return_value=[]):
+        ctx = await mm.enrich_context(FREE_USER_ID, "test message")
+
+    assert ctx.get("relational_memory") == []
+
+
+@pytest.mark.asyncio
+async def test_decay_relations_reduces_confidence(db_session, pro_user_with_key):
+    """decay_relations reduces confidence on stale rows."""
+    old_date = datetime.now(timezone.utc) - timedelta(days=35)
+    row = MemoryRelation(
+        id=str(uuid.uuid4()),
+        user_id=PRO_USER_ID,
+        subject_label="OldContact",
+        subject_type="person",
+        predicate="knows",
+        object_label="SomeProject",
+        object_type="project",
+        confidence=0.8,
+        last_confirmed_at=old_date,
+    )
+    db_session.add(row)
+    await db_session.commit()
+
+    await decay_relations(db_session, PRO_USER_ID)
+
+    result = await db_session.execute(
+        select(MemoryRelation).where(MemoryRelation.subject_label == "OldContact")
+    )
+    updated = result.scalar_one_or_none()
+    assert updated is not None
+    assert updated.confidence < 0.8
+
+
+@pytest.mark.asyncio
+async def test_decay_relations_prunes_low_confidence(db_session, pro_user_with_key):
+    """decay_relations deletes rows whose confidence drops below 0.2 threshold."""
+    # Start at 0.21 with 60-day-old last_confirmed_at → two decay periods → 0.21 * 0.95^2 ≈ 0.19 → pruned
+    old_date = datetime.now(timezone.utc) - timedelta(days=65)
+    row = MemoryRelation(
+        id=str(uuid.uuid4()),
+        user_id=PRO_USER_ID,
+        subject_label="ExpiredContact",
+        subject_type="person",
+        predicate="used_to_work_with",
+        object_label="OldCorp",
+        object_type="company",
+        confidence=0.21,
+        last_confirmed_at=old_date,
+    )
+    db_session.add(row)
+    await db_session.commit()
+
+    await decay_relations(db_session, PRO_USER_ID)
+
+    result = await db_session.execute(
+        select(MemoryRelation).where(MemoryRelation.subject_label == "ExpiredContact")
+    )
+    pruned = result.scalar_one_or_none()
+    assert pruned is None

tests/test_plugins.py

@@ -1,400 +0,0 @@
-"""Tests for Step 10+12: Plugin Marketplace (DB-backed).
-
-Covers:
-  - PluginRegistry: catalog management, filtering, sorting, install counts (PostgreSQL)
-  - ReviewQueue: pending queue, review decisions, manifest security checklist
-  - RevenueShare: install event recording, earnings aggregation (PostgreSQL)
-  - Route integration: tier gate, list/get/install/uninstall via TestClient
-"""
-
-from __future__ import annotations
-
-import uuid
-
-import pytest
-from sqlalchemy import select
-from sqlalchemy.ext.asyncio import AsyncSession
-
-from app.marketplace.plugin_registry import PluginRegistry
-from app.marketplace.plugin_review import ReviewQueue, validate_manifest
-from app.marketplace.revenue_share import RevenueShare
-from app.models import Plugin, PluginReview as PluginReviewModel, RevenueEvent
-from app.schemas import PluginManifest
-from tests.conftest import TEST_USER_IDS, auth_header
-
-# ---------------------------------------------------------------------------
-# Helpers
-# ---------------------------------------------------------------------------
-
-
-def _fresh_manifest(
-    plugin_id: str | None = None,
-    category: str = "productivity",
-    price_cents: int = 0,
-    permissions: list[str] | None = None,
-) -> PluginManifest:
-    pid = plugin_id or f"plugin-{uuid.uuid4().hex[:8]}"
-    return PluginManifest(
-        id=pid,
-        name=f"Plugin {pid}",
-        description=f"Description for {pid}",
-        version="1.0.0",
-        author="test-author",
-        permissions=permissions or ["read:tasks"],
-        category=category,
-        price_cents=price_cents,
-    )
-
-
-# ---------------------------------------------------------------------------
-# PluginRegistry (DB-backed)
-# ---------------------------------------------------------------------------
-
-
-class TestPluginRegistry:
-    """Each test uses the conftest db_session fixture with a fresh in-memory DB."""
-
-    @pytest.fixture
-    def reg(self) -> PluginRegistry:
-        return PluginRegistry()
-
-    @pytest.mark.asyncio
-    async def test_seed_plugins_are_listed(
-        self, reg: PluginRegistry, db_session: AsyncSession, seed_plugins: list[Plugin]
-    ) -> None:
-        result = await reg.list_plugins(db_session)
-        assert result.total == 3
-        assert all(p.id.startswith("plugin-") for p in result.plugins)
-
-    @pytest.mark.asyncio
-    async def test_list_approved_only(
-        self, reg: PluginRegistry, db_session: AsyncSession, seed_plugins: list[Plugin]
-    ) -> None:
-        manifest = _fresh_manifest()
-        await reg.submit_plugin(db_session, manifest, "plugins/key.zip")
-        result = await reg.list_plugins(db_session)
-        ids = [p.id for p in result.plugins]
-        assert manifest.id not in ids  # still pending
-
-    @pytest.mark.asyncio
-    async def test_list_filter_by_category(
-        self, reg: PluginRegistry, db_session: AsyncSession, seed_plugins: list[Plugin]
-    ) -> None:
-        result = await reg.list_plugins(db_session, category="communication")
-        assert result.total == 1
-        assert result.plugins[0].id == "plugin-slack-notify"
-
-    @pytest.mark.asyncio
-    async def test_list_filter_by_query(
-        self, reg: PluginRegistry, db_session: AsyncSession, seed_plugins: list[Plugin]
-    ) -> None:
-        result = await reg.list_plugins(db_session, query="time")
-        assert result.total == 1
-        assert result.plugins[0].id == "plugin-time-tracker"
-
-    @pytest.mark.asyncio
-    async def test_list_sort_by_installs(
-        self, reg: PluginRegistry, db_session: AsyncSession, seed_plugins: list[Plugin]
-    ) -> None:
-        await reg.record_install(db_session, "plugin-slack-notify")
-        await reg.record_install(db_session, "plugin-slack-notify")
-        result = await reg.list_plugins(db_session, sort="installs")
-        assert result.plugins[0].id == "plugin-slack-notify"
-
-    @pytest.mark.asyncio
-    async def test_get_plugin_found(
-        self, reg: PluginRegistry, db_session: AsyncSession, seed_plugins: list[Plugin]
-    ) -> None:
-        entry = await reg.get_plugin(db_session, "plugin-github-sync")
-        assert entry is not None
-        assert entry["manifest"].id == "plugin-github-sync"
-        assert "install_count" in entry
-
-    @pytest.mark.asyncio
-    async def test_get_plugin_not_found(
-        self, reg: PluginRegistry, db_session: AsyncSession
-    ) -> None:
-        entry = await reg.get_plugin(db_session, "no-such-plugin")
-        assert entry is None
-
-    @pytest.mark.asyncio
-    async def test_submit_sets_pending(
-        self, reg: PluginRegistry, db_session: AsyncSession
-    ) -> None:
-        manifest = _fresh_manifest()
-        plugin_id = await reg.submit_plugin(db_session, manifest, "key.zip")
-        assert plugin_id == manifest.id
-        result = await db_session.execute(select(Plugin).where(Plugin.id == plugin_id))
-        row = result.scalar_one()
-        assert row.status == "pending_review"
-
-    @pytest.mark.asyncio
-    async def test_approve_makes_visible(
-        self, reg: PluginRegistry, db_session: AsyncSession
-    ) -> None:
-        manifest = _fresh_manifest()
-        await reg.submit_plugin(db_session, manifest, "key.zip")
-        await reg.approve_plugin(db_session, manifest.id)
-        result = await reg.list_plugins(db_session)
-        assert manifest.id in [p.id for p in result.plugins]
-
-    @pytest.mark.asyncio
-    async def test_reject_stores_reason(
-        self, reg: PluginRegistry, db_session: AsyncSession
-    ) -> None:
-        manifest = _fresh_manifest()
-        await reg.submit_plugin(db_session, manifest, "key.zip")
-        await reg.reject_plugin(db_session, manifest.id, reason="Unsafe permissions")
-        result = await db_session.execute(select(Plugin).where(Plugin.id == manifest.id))
-        row = result.scalar_one()
-        assert row.status == "rejected"
-        assert row.rejection_reason == "Unsafe permissions"
-        listed = await reg.list_plugins(db_session)
-        assert manifest.id not in [p.id for p in listed.plugins]
-
-    @pytest.mark.asyncio
-    async def test_approve_unknown_raises_key_error(
-        self, reg: PluginRegistry, db_session: AsyncSession
-    ) -> None:
-        with pytest.raises(KeyError):
-            await reg.approve_plugin(db_session, "ghost-plugin")
-
-    @pytest.mark.asyncio
-    async def test_record_install_increments_count(
-        self, reg: PluginRegistry, db_session: AsyncSession, seed_plugins: list[Plugin]
-    ) -> None:
-        await reg.record_install(db_session, "plugin-github-sync")
-        entry = await reg.get_plugin(db_session, "plugin-github-sync")
-        assert entry is not None
-        assert entry["install_count"] == 1
-
-    @pytest.mark.asyncio
-    async def test_record_uninstall_decrements_count(
-        self, reg: PluginRegistry, db_session: AsyncSession, seed_plugins: list[Plugin]
-    ) -> None:
-        await reg.record_install(db_session, "plugin-github-sync")
-        await reg.record_install(db_session, "plugin-github-sync")
-        await reg.record_uninstall(db_session, "plugin-github-sync")
-        entry = await reg.get_plugin(db_session, "plugin-github-sync")
-        assert entry is not None
-        assert entry["install_count"] == 1
-
-    @pytest.mark.asyncio
-    async def test_record_uninstall_floors_at_zero(
-        self, reg: PluginRegistry, db_session: AsyncSession, seed_plugins: list[Plugin]
-    ) -> None:
-        await reg.record_uninstall(db_session, "plugin-github-sync")
-        entry = await reg.get_plugin(db_session, "plugin-github-sync")
-        assert entry is not None
-        assert entry["install_count"] == 0
-
-
-# ---------------------------------------------------------------------------
-# ReviewQueue (DB-backed)
-# ---------------------------------------------------------------------------
-
-
-class TestReviewQueue:
-    @pytest.fixture
-    def reg(self) -> PluginRegistry:
-        return PluginRegistry()
-
-    @pytest.fixture
-    def queue(self) -> ReviewQueue:
-        return ReviewQueue()
-
-    @pytest.mark.asyncio
-    async def test_get_pending_returns_submitted_plugins(
-        self, reg: PluginRegistry, queue: ReviewQueue, db_session: AsyncSession
-    ) -> None:
-        manifest = _fresh_manifest()
-        await reg.submit_plugin(db_session, manifest, "key.zip")
-        pending = await queue.get_pending(db_session)
-        assert any(p["plugin_id"] == manifest.id for p in pending)
-
-    @pytest.mark.asyncio
-    async def test_submit_review_approved(
-        self, reg: PluginRegistry, queue: ReviewQueue, db_session: AsyncSession
-    ) -> None:
-        manifest = _fresh_manifest()
-        await reg.submit_plugin(db_session, manifest, "key.zip")
-        await queue.submit_review(db_session, manifest.id, TEST_USER_IDS["power"], "approved", "Looks good")
-        result = await db_session.execute(select(Plugin).where(Plugin.id == manifest.id))
-        row = result.scalar_one()
-        assert row.status == "approved"
-        # Check review row was persisted
-        review_result = await db_session.execute(
-            select(PluginReviewModel).where(PluginReviewModel.plugin_id == manifest.id)
-        )
-        review = review_result.scalar_one()
-        assert review.decision == "approved"
-
-    @pytest.mark.asyncio
-    async def test_submit_review_rejected(
-        self, reg: PluginRegistry, queue: ReviewQueue, db_session: AsyncSession
-    ) -> None:
-        manifest = _fresh_manifest()
-        await reg.submit_plugin(db_session, manifest, "key.zip")
-        await queue.submit_review(
-            db_session, manifest.id, TEST_USER_IDS["power"], "rejected", "Bad permissions"
-        )
-        result = await db_session.execute(select(Plugin).where(Plugin.id == manifest.id))
-        row = result.scalar_one()
-        assert row.status == "rejected"
-
-    def test_validate_manifest_ok(self) -> None:
-        manifest = _fresh_manifest(permissions=["read:tasks", "write:notes"])
-        validate_manifest(manifest)  # should not raise
-
-    def test_validate_manifest_unknown_permission(self) -> None:
-        manifest = _fresh_manifest(permissions=["read:tasks", "read:secrets"])
-        with pytest.raises(ValueError, match="Unknown permission"):
-            validate_manifest(manifest)
-
-    def test_validate_manifest_invalid_id_format(self) -> None:
-        manifest = _fresh_manifest(plugin_id="Plugin_ID_Invalid")
-        with pytest.raises(ValueError, match="Invalid plugin id format"):
-            validate_manifest(manifest)
-
-    def test_validate_manifest_id_with_uppercase(self) -> None:
-        manifest = _fresh_manifest(plugin_id="UpperCase")
-        with pytest.raises(ValueError, match="Invalid plugin id format"):
-            validate_manifest(manifest)
-
-
-# ---------------------------------------------------------------------------
-# RevenueShare (DB-backed)
-# ---------------------------------------------------------------------------
-
-
-class TestRevenueShare:
-    @pytest.fixture
-    def rs(self) -> RevenueShare:
-        return RevenueShare()
-
-    @pytest.mark.asyncio
-    async def test_record_install_free_plugin(
-        self, rs: RevenueShare, db_session: AsyncSession, seed_plugins: list[Plugin]
-    ) -> None:
-        await rs.record_install(db_session, "plugin-github-sync", TEST_USER_IDS["power"], amount_cents=0)
-        result = await db_session.execute(
-            select(RevenueEvent).where(RevenueEvent.plugin_id == "plugin-github-sync")
-        )
-        event = result.scalar_one()
-        assert event.developer_share_cents == 0
-
-    @pytest.mark.asyncio
-    async def test_record_install_paid_plugin_no_stripe(
-        self, rs: RevenueShare, db_session: AsyncSession, seed_plugins: list[Plugin]
-    ) -> None:
-        await rs.record_install(
-            db_session, "plugin-slack-notify", TEST_USER_IDS["pro"], amount_cents=499
-        )
-        result = await db_session.execute(
-            select(RevenueEvent).where(RevenueEvent.plugin_id == "plugin-slack-notify")
-        )
-        event = result.scalar_one()
-        assert event.amount_cents == 499
-        assert event.developer_share_cents == int(499 * 0.70)
-
-    @pytest.mark.asyncio
-    async def test_record_install_increments_registry_count(
-        self, rs: RevenueShare, db_session: AsyncSession, seed_plugins: list[Plugin]
-    ) -> None:
-        reg = PluginRegistry()
-        await rs.record_install(db_session, "plugin-github-sync", TEST_USER_IDS["power"], amount_cents=0)
-        entry = await reg.get_plugin(db_session, "plugin-github-sync")
-        assert entry is not None
-        assert entry["install_count"] == 1
-
-    @pytest.mark.asyncio
-    async def test_get_earnings_empty(
-        self, rs: RevenueShare, db_session: AsyncSession
-    ) -> None:
-        result = await rs.get_earnings(db_session, "unknown-dev")
-        assert result["total_installs"] == 0
-        assert result["total_revenue_cents"] == 0
-        assert result["developer_share_cents"] == 0
-
-    @pytest.mark.asyncio
-    async def test_get_earnings_aggregates(
-        self, rs: RevenueShare, db_session: AsyncSession, seed_plugins: list[Plugin]
-    ) -> None:
-        await rs.record_install(db_session, "plugin-slack-notify", TEST_USER_IDS["power"], amount_cents=499)
-        await rs.record_install(db_session, "plugin-slack-notify", TEST_USER_IDS["pro"], amount_cents=499)
-        result = await rs.get_earnings(db_session, "Adiuva")
-        assert result["total_installs"] == 2
-        assert result["total_revenue_cents"] == 998
-        assert result["developer_share_cents"] == int(499 * 0.70) * 2
-
-
-# ---------------------------------------------------------------------------
-# Route integration tests
-# ---------------------------------------------------------------------------
-
-
-class TestPluginRoutes:
-    def test_list_plugins_requires_power_tier(self, client, seed_plugins) -> None:
-        resp = client.get("/api/v1/plugins", headers=auth_header("free"))
-        assert resp.status_code == 403
-
-    def test_list_plugins_pro_tier_blocked(self, client, seed_plugins) -> None:
-        resp = client.get("/api/v1/plugins", headers=auth_header("pro"))
-        assert resp.status_code == 403
-
-    def test_list_plugins_power_tier_ok(self, client, seed_plugins) -> None:
-        resp = client.get("/api/v1/plugins", headers=auth_header("power"))
-        assert resp.status_code == 200
-        data = resp.json()
-        assert "plugins" in data
-        assert data["total"] == 3
-
-    def test_list_plugins_team_tier_ok(self, client, seed_plugins) -> None:
-        resp = client.get("/api/v1/plugins", headers=auth_header("team"))
-        assert resp.status_code == 200
-
-    def test_get_plugin_found(self, client, seed_plugins) -> None:
-        resp = client.get("/api/v1/plugins/plugin-github-sync", headers=auth_header())
-        assert resp.status_code == 200
-        data = resp.json()
-        assert data["plugin"]["id"] == "plugin-github-sync"
-        assert "install_count" in data
-
-    def test_get_plugin_not_found(self, client, seed_plugins) -> None:
-        resp = client.get("/api/v1/plugins/no-such-plugin", headers=auth_header())
-        assert resp.status_code == 404
-
-    def test_install_plugin_free(self, client, seed_plugins) -> None:
-        resp = client.post(
-            "/api/v1/plugins/plugin-github-sync/install",
-            json={"plugin_id": "plugin-github-sync"},
-            headers=auth_header(),
-        )
-        assert resp.status_code == 200
-        data = resp.json()
-        assert data["ok"] is True
-        assert "download_url" in data
-
-    def test_install_plugin_not_found(self, client, seed_plugins) -> None:
-        resp = client.post(
-            "/api/v1/plugins/ghost/install",
-            json={"plugin_id": "ghost"},
-            headers=auth_header(),
-        )
-        assert resp.status_code == 404
-
-    def test_uninstall_plugin_ok(self, client, seed_plugins) -> None:
-        resp = client.delete(
-            "/api/v1/plugins/plugin-github-sync/install",
-            headers=auth_header(),
-        )
-        assert resp.status_code == 200
-        assert resp.json()["ok"] is True
-
-    def test_install_requires_power_tier(self, client, seed_plugins) -> None:
-        resp = client.post(
-            "/api/v1/plugins/plugin-github-sync/install",
-            json={"plugin_id": "plugin-github-sync"},
-            headers=auth_header("free"),
-        )
-        assert resp.status_code == 403

tests/test_preprocessors.py

@@ -0,0 +1,97 @@
+"""Tests for the preprocessor system (Step 1 — Local Agent V2).
+
+Run:
+    pytest tests/test_preprocessors.py -v
+    pytest tests/test_preprocessors.py -v --preprocess-dir /path/to/folder
+
+The folder must contain cases.yaml + data/.
+"""
+
+from __future__ import annotations
+
+import re
+from pathlib import Path
+
+import yaml
+
+from app.core.preprocessors import detect_content_type, preprocess
+
+_DEFAULT_DIR = Path(__file__).parent / "fixtures" / "preprocessors"
+
+_GENERATORS = {
+    "binary_noise": "some\x00\x01\x02\x03\x04\x05content" * 20,
+}
+
+
+def _fixtures_dir(config) -> Path:
+    override = config.getoption("--preprocess-dir")
+    return Path(override) if override else _DEFAULT_DIR
+
+
+def _load_cases(config) -> list[dict]:
+    return yaml.safe_load((_fixtures_dir(config) / "cases.yaml").read_text(encoding="utf-8"))
+
+
+def _content(case: dict, data_dir: Path) -> str:
+    if "generate" in case:
+        return _GENERATORS[case["generate"]]
+    return (data_dir / case["file"]).read_text(encoding="utf-8")
+
+
+# ── parametrize at collection time via pytest hook ────────────────────
+
+def pytest_generate_tests(metafunc):
+    if "preprocess_case" not in metafunc.fixturenames:
+        return
+    cases = _load_cases(metafunc.config)
+    test_name = metafunc.function.__name__
+    if test_name == "test_detect":
+        subset = [c for c in cases if "detect" in c]
+    else:
+        subset = [c for c in cases if "process" in c]
+    metafunc.parametrize("preprocess_case", subset, ids=[c["id"] for c in subset])
+
+
+# ── detect ────────────────────────────────────────────────────────────
+
+def test_detect(preprocess_case, pytestconfig) -> None:
+    case = preprocess_case
+    data_dir = _fixtures_dir(pytestconfig) / "data"
+    raw = _content(case, data_dir)
+    filename = case.get("file", "")
+    ct = detect_content_type(filename, raw)
+    expected = case["detect"]
+    assert ct == expected, f"[{case['id']}] expected {expected!r}, got {ct!r}"
+
+
+# ── preprocess ────────────────────────────────────────────────────────
+
+def test_preprocess(preprocess_case, pytestconfig) -> None:
+    case = preprocess_case
+    data_dir = _fixtures_dir(pytestconfig) / "data"
+    raw = _content(case, data_dir)
+    result = preprocess(case["process"], raw)
+
+    if case.get("no_html"):
+        assert not re.search(r"<[^>]+>", result.clean_text), "clean_text contains HTML tags"
+
+    if "min_chars" in case:
+        assert len(result.clean_text) >= case["min_chars"], \
+            f"clean_text too short: {len(result.clean_text)} < {case['min_chars']}"
+
+    if "ratio_lt" in case:
+        ratio = len(result.clean_text) / len(raw)
+        assert ratio < case["ratio_lt"], f"compression ratio {ratio:.2f} >= {case['ratio_lt']}"
+
+    for key in case.get("has_meta", []):
+        assert result.metadata.get(key), f"metadata missing {key!r} (got {result.metadata})"
+
+    for item in ([case["contains"]] if isinstance(case.get("contains"), str) else case.get("contains", [])):
+        assert item in result.clean_text, f"clean_text missing {item!r}"
+
+    for item in ([case["excludes"]] if isinstance(case.get("excludes"), str) else case.get("excludes", [])):
+        assert item not in result.clean_text, f"clean_text contains forbidden {item!r}"
+
+    if "content_type" in case:
+        assert result.content_type == case["content_type"], \
+            f"expected content_type {case['content_type']!r}, got {result.content_type!r}"