Files
api/app/scouts/engine.py

271 lines
11 KiB
Python

"""ScoutEngine — orchestrates triage, queueing, and delivery for cloud scouts.
Triage flow per scout:
1. Resolve scout config from the DB.
2. Skip if device hasn't connected within ``device_inactivity_pause_days``.
3. Ask the connector to ``list_new`` — fresh items since last poll.
4. For each item:
- skip if already in the queue (idempotent on (scout_id, source_msg_ref))
- fetch the full content via the connector (transient, never persisted)
- run the triage LLM call → relevant | spam
- spam + auto_trash_spam → connector.archive
- relevant → INSERT scout_triage_queue row
5. Update scout.last_run_at.
Delivery flow on Electron WS reconnect:
- drain ``status='queued'`` rows for the user
- fetch metadata-only for each (subject + snippet)
- send a ``scout_proposal`` frame
- flip status to ``delivered`` on ack
"""
from __future__ import annotations
import logging
import uuid
from datetime import datetime, timedelta, timezone
from sqlalchemy import select
from sqlalchemy.exc import IntegrityError
from app.core.langfuse_client import extract_usage, get_langfuse, get_prompt_or_fallback
from app.core.llm import get_llm
from app.db import async_session
from app.models import CloudScoutConfig, ScoutTriageQueue
from app.scouts.connectors.base import ItemContent, ItemRef, TriageVerdict
from app.scouts.connectors.registry import get_connector
logger = logging.getLogger(__name__)
QUEUE_TTL_DAYS = 30
class ScoutEngine:
def __init__(self, session_factory=None) -> None:
self._session_factory = session_factory or async_session
async def trigger_scout(self, scout_id: uuid.UUID) -> None:
async with self._session_factory() as session:
scout = await session.get(CloudScoutConfig, str(scout_id))
if scout is None:
logger.warning("trigger_scout: no such scout id=%s", scout_id)
return
if not scout.enabled:
return
# Device-inactivity pause check is a simple heuristic on last_run_at —
# the device-online signal lives in the DeviceConnectionManager and is
# consulted at delivery time. For triage, we only check that the
# configured pause threshold isn't suppressing the run.
connector = get_connector(scout.provider)
try:
refs = await connector.list_new(scout)
except Exception:
logger.exception("scout %s: list_new failed", scout.id)
return
for ref in refs:
await self._process_item(session, scout, connector, ref)
scout.last_run_at = datetime.now(tz=timezone.utc)
await session.commit()
async def _process_item(
self,
session,
scout: CloudScoutConfig,
connector,
ref: ItemRef,
) -> None:
# Idempotency check
existing = await session.execute(
select(ScoutTriageQueue.id).where(
ScoutTriageQueue.scout_id == scout.id,
ScoutTriageQueue.source_msg_ref == ref.source_msg_ref,
)
)
if existing.first() is not None:
return
try:
content = await connector.fetch_content(scout, ref)
except Exception:
logger.exception("scout %s: fetch_content failed for %s", scout.id, ref.source_msg_ref)
return
try:
verdict = await self._triage_llm(scout, content)
except Exception:
logger.exception("scout %s: triage_llm failed for %s", scout.id, ref.source_msg_ref)
return
if verdict.verdict == "spam":
if scout.auto_trash_spam:
try:
await connector.archive(scout, ref)
except Exception:
logger.exception("scout %s: archive failed for %s", scout.id, ref.source_msg_ref)
return
now = datetime.now(tz=timezone.utc)
row = ScoutTriageQueue(
id=str(uuid.uuid4()),
user_id=scout.user_id,
scout_id=scout.id,
source_type=connector.source_type,
source_msg_ref=ref.source_msg_ref,
triage_verdict=verdict.verdict,
triage_reason=verdict.reason,
status="queued",
triaged_at=now,
expires_at=now + timedelta(days=QUEUE_TTL_DAYS),
)
session.add(row)
try:
# Use a savepoint so an IntegrityError on race doesn't poison the
# outer session — works on both PostgreSQL (SAVEPOINT) and SQLite.
async with session.begin_nested():
await session.flush()
except IntegrityError:
# Race: another worker inserted between our SELECT and INSERT.
# The unique constraint did its job; safe to ignore.
logger.debug(
"scout %s: idempotent skip for %s (race on unique constraint)",
scout.id,
ref.source_msg_ref,
)
async def deliver_pending(self, user_id: uuid.UUID, ws) -> None:
"""Drain status='queued' rows for user, send scout_proposal WS frames, flip to 'delivered'."""
from app.scouts.connectors.base import ItemRef # noqa: PLC0415
async with self._session_factory() as session:
rows = (await session.execute(
select(ScoutTriageQueue).where(
ScoutTriageQueue.user_id == str(user_id),
ScoutTriageQueue.status == "queued",
)
)).scalars().all()
for row in rows:
try:
connector = get_connector(row.source_type)
except KeyError:
logger.warning("deliver_pending: no connector for %s", row.source_type)
continue
scout = await session.get(CloudScoutConfig, row.scout_id)
if scout is None:
continue
try:
meta = await connector.fetch_metadata(scout, ItemRef(source_msg_ref=row.source_msg_ref))
except Exception:
logger.exception("deliver_pending: fetch_metadata failed")
continue
payload = {
"type": "scout_proposal",
"proposal": {
"id": row.id,
"scout_id": row.scout_id,
"source_type": row.source_type,
"source_msg_ref": row.source_msg_ref,
"raw_subject": meta.subject,
"raw_snippet": meta.snippet,
"category": "unprocessed",
"payload": None,
},
}
await ws.send_json(payload)
row.status = "delivered"
row.delivered_at = datetime.now(tz=timezone.utc)
await session.commit()
async def ack_proposal(self, proposal_id: str) -> None:
"""Flip a delivered proposal to acked. Idempotent — no-op if already acked."""
async with self._session_factory() as session:
row = await session.get(ScoutTriageQueue, proposal_id)
if row is None:
return
row.status = "acked"
row.acked_at = datetime.now(tz=timezone.utc)
await session.commit()
async def _triage_llm(self, scout: CloudScoutConfig, content: ItemContent) -> TriageVerdict:
"""Call the scout-triage-system Langfuse prompt to classify an item as relevant or spam.
Uses gpt-4o-mini with JSON mode. Wraps the LLM call in a Langfuse generation
observation when Langfuse is configured.
"""
import json # noqa: PLC0415
from langchain_core.messages import HumanMessage, SystemMessage # noqa: PLC0415
_TRIAGE_FALLBACK = (
"You are a triage classifier for an executive-assistant scout that watches a "
"{source_type} feed.\n"
'The scout\'s purpose is: "{scout_purpose}".\n\n'
"Given one item, decide whether it is RELEVANT (worth surfacing to the user as a "
"potential task / event / note / project) or SPAM (advertising, mass marketing, "
"phishing, bulk notifications with no actionable content).\n\n"
"Item:\n"
" - Subject: {item_subject}\n"
" - From: {item_sender}\n"
" - Body (truncated): {item_body_truncated_2k}\n\n"
'Return JSON only, matching this schema:\n'
' {{"verdict": "relevant" | "spam", "reason": <short string>, "confidence": <0..1>}}\n\n'
"Be conservative on \"spam\" — if a message could plausibly be a personal/work "
"email, mark it relevant."
)
template, prompt_obj = get_prompt_or_fallback("scout-triage-system", _TRIAGE_FALLBACK)
body_trunc = (content.body_text or "")[:2000]
variables = dict(
source_type=scout.provider,
scout_purpose=scout.prompt_template or "",
item_subject=content.metadata.subject or "",
item_sender=content.metadata.sender or "",
item_body_truncated_2k=body_trunc,
)
if prompt_obj is not None:
try:
system_text = prompt_obj.compile(**variables)
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("scout triage: compile failed: %s", exc)
system_text = template.replace("{{source_type}}", variables["source_type"]) \
.replace("{{scout_purpose}}", variables["scout_purpose"]) \
.replace("{{item_subject}}", variables["item_subject"]) \
.replace("{{item_sender}}", variables["item_sender"]) \
.replace("{{item_body_truncated_2k}}", variables["item_body_truncated_2k"])
else:
system_text = template.format(**variables)
llm = get_llm(model="gpt-4o-mini", temperature=0)
llm_json = llm.bind(response_format={"type": "json_object"}) # type: ignore[attr-defined]
messages = [
SystemMessage(content=system_text),
HumanMessage(content="Classify this item."),
]
lf = get_langfuse()
if lf:
with lf.start_as_current_observation(
as_type="generation",
name="scout-triage",
model="gpt-4o-mini",
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)
data = json.loads(response.content)
return TriageVerdict(**data)