Add web search knowledge base, message tagging, and word-memory tip topics

Introduces websearch.py (Tavily) + knowledge.py to ground chat responses
in search results, tagging.py to tag user and assistant messages via LLM
and pick related topics for the Discord useless-tips job, and
word_memory.py to track recent conversation nouns (surfaced in the new
WordPool.vue sidebar) used to pick tip topics close to a related tag.

backend/app/api/chat.py

@@ -1,3 +1,4 @@
+import asyncio
 import uuid
 from datetime import datetime, timezone
 from fastapi import APIRouter, Depends, HTTPException
@@ -9,7 +10,11 @@ from app.db.database import get_db
 from app.db.models import Session, Message
 from app.core import memory as mem
 from app.core import llm
+from app.core import knowledge
+from app.core.websearch import WEB_SEARCH_TOOL
 from app.core.config import DEFAULT_MODEL
+from app.core.tagging import tag_message_by_id
+from app.core.word_memory import record_message
 
 router = APIRouter(prefix="/chat", tags=["chat"])
 
@@ -35,6 +40,7 @@ class MessageResponse(BaseModel):
     role: str
     content: str
     created_at: datetime
+    tags: list[str] = []
 
     class Config:
         from_attributes = True
@@ -50,7 +56,16 @@ def get_messages(session_id: str, db: DBSession = Depends(get_db)):
     session = db.get(Session, session_id)
     if not session:
         raise HTTPException(status_code=404, detail="Session not found")
-    return session.messages
+    return [
+        MessageResponse(
+            id=m.id,
+            role=m.role,
+            content=m.content,
+            created_at=m.created_at,
+            tags=[t.tag for t in m.tags],
+        )
+        for m in session.messages
+    ]
 
 
 @router.delete("/sessions/{session_id}", status_code=204)
@@ -82,6 +97,10 @@ async def stream_chat(req: ChatRequest, db: DBSession = Depends(get_db)):
     user_msg = Message(session_id=session.id, role="user", content=req.message)
     db.add(user_msg)
     db.commit()
+    db.refresh(user_msg)
+
+    record_message(req.message)
+    asyncio.create_task(tag_message_by_id(user_msg.id))
 
     context = mem.get_context_messages(db, session.id)
 
@@ -94,17 +113,40 @@ async def stream_chat(req: ChatRequest, db: DBSession = Depends(get_db)):
         # セッションIDをまず返す（新規セッション時にフロントへ通知）
         yield sse({"type": "session_id", "value": session.id})
 
+        used_model = req.model or session.model
         full_response = []
         try:
-            async for chunk in llm.stream_chat(context, model=req.model or session.model, session_id=session.id):
+            resolved_context, searches = await llm.resolve_web_search(
+                context, model=used_model, session_id=session.id,
+            )
+            if searches:
+                sources = [
+                    {"title": r["title"], "url": r["url"]}
+                    for s in searches for r in s["results"]
+                ]
+                yield sse({"type": "sources", "value": sources})
+                for s in searches:
+                    summary = "\n".join(r["content"][:300] for r in s["results"] if r["content"])
+                    if summary:
+                        knowledge.save_knowledge(db, s["query"], summary, s["results"])
+
+            search_tools = [WEB_SEARCH_TOOL] if len(resolved_context) > len(context) else None
+            async for chunk in llm.stream_chat(
+                resolved_context, model=used_model, session_id=session.id, tools=search_tools,
+            ):
                 full_response.append(chunk)
                 yield sse({"type": "chunk", "value": chunk})
 
             # アシスタントメッセージを保存
             assistant_content = "".join(full_response)
-            db.add(Message(session_id=session.id, role="assistant", content=assistant_content))
+            assistant_msg = Message(session_id=session.id, role="assistant", content=assistant_content)
+            db.add(assistant_msg)
             session.updated_at = datetime.now(timezone.utc)
             db.commit()
+            db.refresh(assistant_msg)
+
+            record_message(assistant_content)
+            asyncio.create_task(tag_message_by_id(assistant_msg.id))
 
             # メモリ要約チェック
             await mem.maybe_summarize(db, session)

backend/app/api/system.py

@@ -1,9 +1,13 @@
-from fastapi import APIRouter
+from fastapi import APIRouter, Depends
 import psutil
 from pydantic import BaseModel
+from sqlalchemy.orm import Session as DBSession
 
 from app.core.scheduler import send_useless_tip, send_weather_laundry_notification
+from app.core.tagging import backfill_tags
 from app.core.weather_cache import get_latest as get_latest_weather
+from app.core.word_memory import get_word_pool_with_counts
+from app.db.database import get_db
 
 router = APIRouter(prefix="/api/system", tags=["system"])
 
@@ -68,3 +72,16 @@ async def test_useless_tip():
 async def get_latest_weather_data():
     """直近で取得した明日の天気データを返す（未取得ならnull）"""
     return get_latest_weather()
+
+
+@router.get("/word-pool")
+async def get_word_pool():
+    """どうでもいいtips生成に使われているワードプールを出現回数順で返す"""
+    return get_word_pool_with_counts()
+
+
+@router.post("/tags/backfill")
+async def run_tags_backfill(limit: int | None = None, db: DBSession = Depends(get_db)):
+    """過去のユーザーメッセージのうち、タグが未生成のものにまとめてタグを付ける"""
+    count = await backfill_tags(db, limit=limit)
+    return {"tagged_count": count}