Re-embed the knowledge base when the embedding model changes

Vectors from two different embedding models are not comparable — Titan v2 has
1024 dimensions where nomic-embed-text has 768 — and _cosine_similarity zips
them, so a mismatch produced quietly meaningless scores rather than an error.
Nothing forced a re-embed either: sync_local_kb only looked at the content
hash, so switching models left every cached vector in place.

local_knowledge_docs now records the model each vector came from, and a row is
re-embedded when its content hash OR its model differs. Saving a new
LOCAL_KB_EMBEDDING_MODEL from the settings panel starts a background pass over
every file; the knowledge base panel also has a button for it. While that pass
runs, the per-request sync steps aside so no file is embedded twice.

The panel now shows what the embedding model is doing: model name, vector
size, how many files are embedded with the current model, how many are still
on an older one, a progress bar while a pass runs, and the last error. A
"接続テスト" button embeds one short string so a broken or unreachable model
can be told apart from one that is merely slow to warm up.

Also read TAVILY_API_KEY from os.environ at call time. It was imported as a
config constant, which is evaluated before load_into_env() runs, so a key
saved from the settings panel was stored but never used — not even after a
restart.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

backend/app/api/knowledge_base.py

@@ -9,6 +9,7 @@ from fastapi import APIRouter, Depends, HTTPException, Query
 from pydantic import BaseModel
 from sqlalchemy.orm import Session as DBSession
 
+from app.core.memory_kb import local_knowledge
 from app.core.memory_kb.local_knowledge import list_kb_files, read_kb_file
 from app.db.database import get_db
 from app.db.models import LocalKnowledgeDoc
@@ -41,10 +42,14 @@ class KbFileContent(BaseModel):
 @router.get("/files", response_model=list[KbFileInfo])
 def list_files(db: DBSession = Depends(get_db)):
     """ナレッジベース配下の Markdown を一覧で返す（本文は含めず、先頭のプレビューのみ）。"""
+    # 「埋め込み済み」は現在のモデルで埋め込まれている行だけを指す（別モデルの行は埋め直し待ち）
+    current_model = local_knowledge.current_embedding_model()
     indexed_paths = {
         # DB 側は OS 依存の区切り文字で保存されているため、比較用に posix 形式へ寄せる
         path.replace("\\", "/")
-        for (path,) in db.query(LocalKnowledgeDoc.path).all()
+        for (path,) in db.query(LocalKnowledgeDoc.path)
+        .filter(LocalKnowledgeDoc.embedding_model == current_model)
+        .all()
     }
     return [
         KbFileInfo(**info, indexed=info["path"] in indexed_paths)
@@ -59,3 +64,49 @@ def get_file(path: str = Query(..., description="ナレッジベースディレ
     if file is None:
         raise HTTPException(status_code=404, detail="Knowledge base file not found")
     return KbFileContent(**file)
+
+
+class EmbeddingStatus(BaseModel):
+    """埋め込みモデルの現状。GUI の状態表示・進捗バーに使う。"""
+    model: str
+    # 現在のモデルで埋め込まれたベクトルの次元数（1件も無ければ None）
+    dimensions: int | None
+    file_count: int
+    indexed: int
+    # 別のモデルで埋め込まれていて埋め直しが必要な件数
+    stale: int
+    running: bool
+    total: int
+    done: int
+    failed: int
+    started_at: datetime | None
+    finished_at: datetime | None
+    last_error: str | None
+
+
+class EmbeddingProbe(BaseModel):
+    """短いテキストを実際に埋め込んでみた結果。"""
+    ok: bool
+    model: str
+    dimensions: int | None
+    latency_ms: float
+    error: str | None
+
+
+@router.get("/embedding", response_model=EmbeddingStatus)
+def get_embedding_status(db: DBSession = Depends(get_db)):
+    """現在の埋め込みモデルと、埋め込み済み件数・実行中の進捗を返す。"""
+    return local_knowledge.embedding_status(db)
+
+
+@router.post("/embedding/test", response_model=EmbeddingProbe)
+async def test_embedding():
+    """埋め込みモデルに実際に1回リクエストして、応答するかどうかを確かめる。"""
+    return await local_knowledge.probe_embedding()
+
+
+@router.post("/embedding/reindex", response_model=EmbeddingStatus)
+async def reindex_embeddings(db: DBSession = Depends(get_db)):
+    """全ファイルの再埋め込みをバックグラウンドで開始する。実行中なら何もしない（進捗をそのまま返す）。"""
+    local_knowledge.start_reembed_all()
+    return local_knowledge.embedding_status(db)

backend/app/api/settings.py

@@ -1,9 +1,12 @@
+import os
+
 from fastapi import APIRouter, Depends, HTTPException
 from pydantic import BaseModel
 from sqlalchemy.orm import Session as DBSession
 
 from app.core import settings_store
 from app.core.auth import require_settings_password
+from app.core.memory_kb import local_knowledge
 from app.db.database import get_db
 
 router = APIRouter(prefix="/api/settings", tags=["settings"], dependencies=[Depends(require_settings_password)])
@@ -29,7 +32,8 @@ def list_settings(db: DBSession = Depends(get_db)):
 
 
 @router.put("", response_model=list[SettingEntry])
-def save_settings(req: SaveSettingsRequest, db: DBSession = Depends(get_db)):
+async def save_settings(req: SaveSettingsRequest, db: DBSession = Depends(get_db)):
+    before_embedding_model = os.environ.get("LOCAL_KB_EMBEDDING_MODEL")
     try:
         for key, value in req.values.items():
             if not value:
@@ -37,13 +41,29 @@ def save_settings(req: SaveSettingsRequest, db: DBSession = Depends(get_db)):
             settings_store.save(db, key, value)
     except ValueError as e:
         raise HTTPException(status_code=400, detail=str(e)) from e
+
+    _reembed_if_model_changed(before_embedding_model)
     return settings_store.list_masked(db)
 
 
 @router.delete("/{key}", response_model=list[SettingEntry])
-def clear_setting(key: str, db: DBSession = Depends(get_db)):
+async def clear_setting(key: str, db: DBSession = Depends(get_db)):
+    before_embedding_model = os.environ.get("LOCAL_KB_EMBEDDING_MODEL")
     try:
         settings_store.clear(db, key)
     except ValueError as e:
         raise HTTPException(status_code=400, detail=str(e)) from e
+
+    _reembed_if_model_changed(before_embedding_model)
     return settings_store.list_masked(db)
+
+
+def _reembed_if_model_changed(before: str | None) -> None:
+    """埋め込みモデルが変わったら、ローカルKBの再埋め込みをバックグラウンドで走らせる。
+
+    モデルが違うと次元数も意味空間も変わり、古いベクトルは比較しても無意味なので、
+    設定を変えた時点で全件やり直す（完了を待たずに応答は返す）。
+    """
+    if os.environ.get("LOCAL_KB_EMBEDDING_MODEL") == before:
+        return
+    local_knowledge.start_reembed_all()