Update note management from db vector to index

app/agents/note_agent.py

@@ -1,13 +1,14 @@
-"""Note agent — Markdown note management (list, get, create, update, delete)."""
+"""Note agent — Markdown note management (list, get, create, update, propose edit)."""
 
 from __future__ import annotations
 
+import asyncio
 import re
 from typing import Any
 
 from langchain_core.tools import tool
 
-from app.core.llm import embed
+from app.core.note_summarizer import generate_note_summary
 from app.core.ws_context import execute_on_client
 
 _UUID_RE = re.compile(
@@ -19,9 +20,21 @@ def _is_uuid(value: str) -> bool:
     return bool(_UUID_RE.match(value))
 
 
+def _fmt_summary(row: dict) -> str:
+    summary = (row.get("aiSummary") or row.get("ai_summary") or "").strip()
+    if summary:
+        return f" — {summary}"
+    snippet = (row.get("content") or "")[:120].replace("\n", " ").strip()
+    return f" — {snippet}" if snippet else ""
+
+
 @tool
 async def list_notes(project_id: str = "") -> str:
-    """List notes, optionally scoped to a project by project_id."""
+    """List notes with AI summaries, optionally scoped to a project by project_id.
+
+    Returns id, title, and ai_summary for each note so you can decide which
+    note to read in full with get_note before creating or updating.
+    """
     normalized_project_id = project_id if (project_id and _is_uuid(project_id)) else ""
     result = await execute_on_client(
         action="select",
@@ -31,7 +44,7 @@ async def list_notes(project_id: str = "") -> str:
     rows = result.get("rows", [])
     if not rows:
         return "No notes found."
-    lines = [f"- {r['title']} (id: {r['id']})" for r in rows]
+    lines = [f"  - [{r['id']}] {r['title']}{_fmt_summary(r)}" for r in rows]
     return f"Found {len(rows)} note(s):\n" + "\n".join(lines)
 
 
@@ -66,14 +79,10 @@ async def create_note(
         },
     )
     row = result["row"]
-    # Index the note content in the vector store.
-    vector = await embed(content)
-    await execute_on_client(
-        action="vector_upsert",
-        data={"id": row["id"], "projectId": row.get("projectId"), "content": content},
-        vector=vector,
-    )
-    return f"Note created: '{row['title']}' (id: {row['id']})."
+    note_id: str = row["id"]
+    # Generate summary asynchronously — fire-and-forget.
+    asyncio.create_task(_refresh_summary(note_id, title, content))
+    return f"Note created: '{row['title']}' (id: {note_id})."
 
 
 @tool
@@ -82,7 +91,8 @@ async def update_note(
     title: str = "",
     content: str = "",
 ) -> str:
-    """Update an existing note. Only pass fields that should change.
+    """Update an existing note directly (no approval required).
+    Use propose_note_edit instead when human review is needed.
     note_id: UUID of the note (required)
     If you need to preserve existing content, call get_note first.
     """
@@ -97,17 +107,63 @@ async def update_note(
         data={"id": note_id, "updates": updates},
     )
     row = result["row"]
-    # Re-index if content changed.
     if content:
-        vector = await embed(content)
-        await execute_on_client(
-            action="vector_upsert",
-            data={"id": note_id, "projectId": row.get("projectId"), "content": content},
-            vector=vector,
-        )
+        new_title = title or row.get("title", "")
+        asyncio.create_task(_refresh_summary(note_id, new_title, content))
     return f"Note updated: '{row['title']}' (id: {row['id']})."
 
 
+@tool
+async def propose_note_edit(
+    note_id: str,
+    edit_type: str,
+    proposed_content: str,
+    reasoning: str = "",
+    anchor_before: str = "",
+    anchor_text: str = "",
+    agent_id: str = "",
+    run_id: str = "",
+) -> str:
+    """Propose an AI edit to an existing note, pending human approval.
+
+    Use this instead of update_note when review_required is true.
+    The user will see the proposal highlighted before it is merged.
+
+    note_id: UUID of the target note (required)
+    edit_type: 'append' | 'insert' | 'replace'
+      - append: adds proposed_content at the end of the note
+      - insert: inserts proposed_content immediately after anchor_before text
+      - replace: replaces the first occurrence of anchor_text with proposed_content
+    proposed_content: the new Markdown text to add or substitute (required)
+    reasoning: brief explanation shown to the user (recommended)
+    anchor_before: for 'insert' — the text snippet that precedes the insertion point
+    anchor_text: for 'replace' — the exact text to be replaced
+    agent_id: agent identifier (for traceability)
+    run_id: run identifier (for traceability)
+    """
+    if edit_type not in ("append", "insert", "replace"):
+        return f"Invalid edit_type '{edit_type}'. Use 'append', 'insert', or 'replace'."
+
+    result = await execute_on_client(
+        action="propose_note_edit",
+        data={
+            "noteId": note_id,
+            "type": edit_type,
+            "proposedContent": proposed_content,
+            "reasoning": reasoning or None,
+            "anchorBefore": anchor_before or None,
+            "anchorText": anchor_text or None,
+            "agentId": agent_id or None,
+            "runId": run_id or None,
+        },
+    )
+    edit_id = result.get("id", "?")
+    return (
+        f"Edit proposal created (id: {edit_id}) for note {note_id}. "
+        f"Status: pending user approval."
+    )
+
+
 @tool
 async def delete_note(note_id: str) -> str:
     """Delete a note permanently by its UUID."""
@@ -115,11 +171,32 @@ async def delete_note(note_id: str) -> str:
     return f"Note {note_id} deleted."
 
 
+async def _refresh_summary(note_id: str, title: str, content: str) -> None:
+    """Generate and persist the AI summary for a note.  Fire-and-forget."""
+    try:
+        summary = await generate_note_summary(title, content)
+        if summary:
+            await execute_on_client(
+                action="update",
+                table="notes",
+                data={
+                    "id": note_id,
+                    "updates": {
+                        "aiSummary": summary,
+                        "aiSummaryUpdatedAt": int(__import__("time").time() * 1000),
+                    },
+                },
+            )
+    except Exception:
+        pass  # fire-and-forget; errors logged by generate_note_summary
+
+
 NOTE_TOOLS: list[Any] = [
     list_notes,
     get_note,
     create_note,
     update_note,
+    propose_note_edit,
     delete_note,
 ]
 

app/api/routes/agents.py

@@ -20,10 +20,13 @@ from fastapi import APIRouter, Depends, HTTPException, status
 from sqlalchemy import func, select
 from sqlalchemy.ext.asyncio import AsyncSession
 
+from pydantic import BaseModel
+
 from app.api.deps import get_current_user
 from app.billing.tier_manager import FEATURES
 from app.core.agent_runner import is_agent_running, run_local_agent
 from app.core.device_manager import device_manager
+from app.core.note_summarizer import generate_note_summary
 from app.db import get_session
 from app.models import AgentRunLog, LocalAgentConfig
 from app.schemas import (
@@ -230,3 +233,25 @@ async def trigger_agent_run(
     )
 
     return _to_run_log_response(run_log)
+
+
+# ── Note summary endpoint ──────────────────────────────────────────────────────
+
+
+class NoteSummarizeRequest(BaseModel):
+    title: str
+    content: str
+
+
+class NoteSummarizeResponse(BaseModel):
+    summary: str
+
+
+@router.post("/notes/summarize", response_model=NoteSummarizeResponse)
+async def summarize_note(
+    body: NoteSummarizeRequest,
+    current_user: UserProfile = Depends(get_current_user),
+) -> NoteSummarizeResponse:
+    """Generate an AI summary for a note.  Used by the Electron backfill on startup."""
+    summary = await generate_note_summary(body.title, body.content)
+    return NoteSummarizeResponse(summary=summary)

app/core/agent_runner.py

@@ -658,9 +658,14 @@ async def run_local_agent(
                 # ── Phase B: single LLM call ─────────────────────────
                 extraction_rules = _get_extraction_rules(agent_config, content_type)
                 no_match_behavior = _get_no_match_behavior(agent_config)
-                global_rules_lines = "\n".join(
-                    f"- {r}" for r in agent_config.get("global_rules", [])
-                )
+                base_global_rules = list(agent_config.get("global_rules", []))
+                if "notes" in config.data_types:
+                    base_global_rules.append(
+                        "For notes: when updating an existing note use `propose_note_edit` "
+                        "(type=append/insert/replace) so the user can review AI changes. "
+                        "Only call `update_note` for complete content replacement without review."
+                    )
+                global_rules_lines = "\n".join(f"- {r}" for r in base_global_rules)
                 metadata_section = _format_metadata(preprocessed.metadata)
 
                 system_prompt = compile_prompt(

app/core/llm.py

@@ -111,6 +111,7 @@ _AGENT_MODEL_SETTINGS: dict[str, Callable[[], str]] = {
     "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,
+    "note-summarizer":     lambda: "gpt-4o-mini",
 }
 
 

app/core/note_summarizer.py

@@ -0,0 +1,51 @@
+"""Note summarizer — generates a compact AI summary for a note.
+
+Called fire-and-forget from create_note / update_note tools so the
+``notes.ai_summary`` column stays current without blocking the agent loop.
+"""
+
+from __future__ import annotations
+
+import logging
+
+from langchain_core.messages import HumanMessage, SystemMessage
+
+from app.core.langfuse_client import get_prompt_or_fallback
+from app.core.llm import get_agent_llm
+
+logger = logging.getLogger(__name__)
+
+_FALLBACK_PROMPT = """\
+Summarize this note in <=250 characters. Be terse and dense.
+Keep proper nouns, dates, decisions, and action items.
+Do not start with "This note".
+Respond with the summary text only — no intro, no labels.
+
+Title: {title}
+Content: {content}"""
+
+_MAX_CONTENT_CHARS = 4000
+
+
+async def generate_note_summary(title: str, content: str) -> str:
+    """Return a <=250-char summary of *title* + *content*.
+
+    Uses the Langfuse ``note_summary`` prompt (hot-swappable) with a local
+    fallback.  Truncates *content* to 4000 chars before sending to avoid
+    token waste on large notes.
+    """
+    template, _ = get_prompt_or_fallback("note_summary", _FALLBACK_PROMPT)
+    trimmed = content[:_MAX_CONTENT_CHARS]
+    system_prompt = template.format(title=title, content=trimmed)
+
+    try:
+        llm = get_agent_llm("note-summarizer")
+        response = await llm.ainvoke([
+            SystemMessage(content=system_prompt),
+            HumanMessage(content="Generate the summary."),
+        ])
+        text = response.content if isinstance(response.content, str) else ""
+        return text.strip()[:250]
+    except Exception as exc:
+        logger.warning("note_summarizer: failed to generate summary: %s", exc)
+        return ""