Files
playchat/backend/app/db/models.py
colcaroliandClaude Opus 5 97568ba90e 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>
2026-08-23 22:56:12 +09:00

193 lines
11 KiB
Python

from datetime import datetime, timezone
from sqlalchemy import String, Text, ForeignKey, Integer, Float, UniqueConstraint
from sqlalchemy.orm import Mapped, mapped_column, relationship
from app.db.database import Base
from app.db.types import UtcDateTime
class Session(Base):
__tablename__ = "sessions"
id: Mapped[str] = mapped_column(String, primary_key=True)
title: Mapped[str] = mapped_column(String, default="New Chat")
model: Mapped[str] = mapped_column(String, default="")
created_at: Mapped[datetime] = mapped_column(UtcDateTime, default=lambda: datetime.now(timezone.utc))
updated_at: Mapped[datetime] = mapped_column(UtcDateTime, default=lambda: datetime.now(timezone.utc), onupdate=lambda: datetime.now(timezone.utc))
messages: Mapped[list["Message"]] = relationship(back_populates="session", cascade="all, delete-orphan", order_by="Message.created_at")
memories: Mapped[list["Memory"]] = relationship(back_populates="session", cascade="all, delete-orphan")
note_links: Mapped[list["SessionNotePage"]] = relationship(back_populates="session", cascade="all, delete-orphan")
class Message(Base):
__tablename__ = "messages"
id: Mapped[int] = mapped_column(Integer, primary_key=True, autoincrement=True)
session_id: Mapped[str] = mapped_column(ForeignKey("sessions.id", ondelete="CASCADE"))
role: Mapped[str] = mapped_column(String) # "user" | "assistant" | "system"
content: Mapped[str] = mapped_column(Text)
created_at: Mapped[datetime] = mapped_column(UtcDateTime, default=lambda: datetime.now(timezone.utc))
# 要約されて長期メモリに移ったメッセージ。LLMに渡すコンテキストからは外れるが、
# 人間が後から履歴を確認できるようDBには残す(app.core.memory_kb.memory.maybe_summarize を参照)。
archived_at: Mapped[datetime | None] = mapped_column(UtcDateTime, nullable=True)
# どの要約に取り込まれたか。要約を削除してもメッセージ自体は残す(SET NULL)。
memory_id: Mapped[int | None] = mapped_column(
ForeignKey("memories.id", ondelete="SET NULL"), nullable=True,
)
session: Mapped["Session"] = relationship(back_populates="messages")
tags: Mapped[list["MessageTag"]] = relationship(back_populates="message", cascade="all, delete-orphan")
class MessageTag(Base):
__tablename__ = "message_tags"
id: Mapped[int] = mapped_column(Integer, primary_key=True, autoincrement=True)
message_id: Mapped[int] = mapped_column(ForeignKey("messages.id", ondelete="CASCADE"))
tag: Mapped[str] = mapped_column(String)
created_at: Mapped[datetime] = mapped_column(UtcDateTime, default=lambda: datetime.now(timezone.utc))
message: Mapped["Message"] = relationship(back_populates="tags")
class Memory(Base):
__tablename__ = "memories"
id: Mapped[int] = mapped_column(Integer, primary_key=True, autoincrement=True)
session_id: Mapped[str] = mapped_column(ForeignKey("sessions.id", ondelete="CASCADE"))
summary: Mapped[str] = mapped_column(Text)
embedding_json: Mapped[str | None] = mapped_column(Text, nullable=True) # 将来のベクトル検索用
created_at: Mapped[datetime] = mapped_column(UtcDateTime, default=lambda: datetime.now(timezone.utc))
session: Mapped["Session"] = relationship(back_populates="memories")
tags: Mapped[list["MemoryTag"]] = relationship(back_populates="memory", cascade="all, delete-orphan")
class MemoryTag(Base):
__tablename__ = "memory_tags"
id: Mapped[int] = mapped_column(Integer, primary_key=True, autoincrement=True)
memory_id: Mapped[int] = mapped_column(ForeignKey("memories.id", ondelete="CASCADE"))
tag: Mapped[str] = mapped_column(String)
created_at: Mapped[datetime] = mapped_column(UtcDateTime, default=lambda: datetime.now(timezone.utc))
memory: Mapped["Memory"] = relationship(back_populates="tags")
class KnowledgeEntry(Base):
"""Web検索で得た知識。セッションを横断して蓄積し、以後の会話のシステムプロンプトに注入する。"""
__tablename__ = "knowledge_entries"
id: Mapped[int] = mapped_column(Integer, primary_key=True, autoincrement=True)
query: Mapped[str] = mapped_column(String)
summary: Mapped[str] = mapped_column(Text)
# システムプロンプトには本文ではなくこの一行要約だけを並べる(app.core.memory_kb.knowledge を参照)。
# このカラム追加より前に作られた行では None になるため、表示側で summary から導出する。
headline: Mapped[str | None] = mapped_column(String, nullable=True)
created_at: Mapped[datetime] = mapped_column(UtcDateTime, default=lambda: datetime.now(timezone.utc))
sources: Mapped[list["KnowledgeSource"]] = relationship(back_populates="entry", cascade="all, delete-orphan")
class KnowledgeSource(Base):
__tablename__ = "knowledge_sources"
id: Mapped[int] = mapped_column(Integer, primary_key=True, autoincrement=True)
entry_id: Mapped[int] = mapped_column(ForeignKey("knowledge_entries.id", ondelete="CASCADE"))
title: Mapped[str] = mapped_column(String)
url: Mapped[str] = mapped_column(String)
entry: Mapped["KnowledgeEntry"] = relationship(back_populates="sources")
class LocalKnowledgeDoc(Base):
"""backend/knowledge_base/ 配下のMarkdownファイルとその埋め込みベクトル。
話題に応じて関連ファイルをシステムプロンプトに注入するために使う。
ファイル内容のハッシュか埋め込みモデルが変わった時だけ埋め込みを再計算する。
"""
__tablename__ = "local_knowledge_docs"
id: Mapped[int] = mapped_column(Integer, primary_key=True, autoincrement=True)
path: Mapped[str] = mapped_column(String, unique=True)
title: Mapped[str] = mapped_column(String)
content_hash: Mapped[str] = mapped_column(String)
embedding_json: Mapped[str] = mapped_column(Text)
# この埋め込みを計算したモデル名。モデルを切り替えた行を見分けて埋め直すために持つ
# (次元数の違うベクトル同士を比較しても意味のないスコアしか出ないため)。
# このカラムより前に作られた行は NULL で、モデル不明として埋め直し対象になる。
embedding_model: Mapped[str | None] = mapped_column(String, nullable=True)
updated_at: Mapped[datetime] = mapped_column(UtcDateTime, default=lambda: datetime.now(timezone.utc), onupdate=lambda: datetime.now(timezone.utc))
class NotePage(Base):
"""ユーザーとAIが共同で編集する共有ノートの1ページ。UIではタブとして表示する。
ページのタイトル一覧は毎回システムプロンプトに注入され、本文はモデルが
read_note ツールで読み出す(app.core.notes を参照)。ただし SessionNotePage で
セッションに紐付けられたページは、本文もそのままシステムプロンプトに載る。
"""
__tablename__ = "note_pages"
id: Mapped[int] = mapped_column(Integer, primary_key=True, autoincrement=True)
title: Mapped[str] = mapped_column(String)
content: Mapped[str] = mapped_column(Text, default="")
position: Mapped[int] = mapped_column(Integer, default=0) # タブの並び順(セクションでまとめる際、同じセクションのページは連続させる)
section: Mapped[str | None] = mapped_column(String, nullable=True) # タブをグループ化するセクション名。未設定なら未分類
created_at: Mapped[datetime] = mapped_column(UtcDateTime, default=lambda: datetime.now(timezone.utc))
updated_at: Mapped[datetime] = mapped_column(UtcDateTime, default=lambda: datetime.now(timezone.utc), onupdate=lambda: datetime.now(timezone.utc))
session_links: Mapped[list["SessionNotePage"]] = relationship(back_populates="page", cascade="all, delete-orphan")
class SessionNotePage(Base):
"""「このセッションではこのページの本文をプロンプトに載せる」という指定。
通常、共有ノートはタイトル一覧だけがシステムプロンプトに入り、本文はモデルが
read_note ツールで取りに行く。ここに登録されたページはそれを待たず、本文が毎回
そのままシステムプロンプトに入る(app.core.notes.get_session_notes_context を参照)。
セッション単位なので、会話ごとに違う組み合わせを選べる。
"""
__tablename__ = "session_note_pages"
__table_args__ = (UniqueConstraint("session_id", "page_id", name="uq_session_note_page"),)
id: Mapped[int] = mapped_column(Integer, primary_key=True, autoincrement=True)
session_id: Mapped[str] = mapped_column(ForeignKey("sessions.id", ondelete="CASCADE"))
page_id: Mapped[int] = mapped_column(ForeignKey("note_pages.id", ondelete="CASCADE"))
created_at: Mapped[datetime] = mapped_column(UtcDateTime, default=lambda: datetime.now(timezone.utc))
session: Mapped["Session"] = relationship(back_populates="note_links")
page: Mapped["NotePage"] = relationship(back_populates="session_links")
class AppSetting(Base):
"""GUIの設定画面から編集する接続情報(APIキー等)。app.core.settings_store が読み書きする。"""
__tablename__ = "app_settings"
key: Mapped[str] = mapped_column(String, primary_key=True)
value: Mapped[str] = mapped_column(Text)
updated_at: Mapped[datetime] = mapped_column(UtcDateTime, default=lambda: datetime.now(timezone.utc), onupdate=lambda: datetime.now(timezone.utc))
class ModelPreset(Base):
"""モデル選択プルダウン(ModelSelector.vue)に表示するプリセット。GUIの設定画面から編集する。"""
__tablename__ = "model_presets"
id: Mapped[int] = mapped_column(Integer, primary_key=True, autoincrement=True)
label: Mapped[str] = mapped_column(String)
value: Mapped[str] = mapped_column(String, unique=True)
created_at: Mapped[datetime] = mapped_column(UtcDateTime, default=lambda: datetime.now(timezone.utc))
class ApiLog(Base):
__tablename__ = "api_logs"
id: Mapped[int] = mapped_column(Integer, primary_key=True, autoincrement=True)
session_id: Mapped[str | None] = mapped_column(ForeignKey("sessions.id", ondelete="SET NULL"), nullable=True)
model: Mapped[str] = mapped_column(String)
request_json: Mapped[str] = mapped_column(Text) # JSON: messages array
response_text: Mapped[str | None] = mapped_column(Text, nullable=True)
error: Mapped[str | None] = mapped_column(Text, nullable=True)
latency_ms: Mapped[float | None] = mapped_column(Float, nullable=True)
created_at: Mapped[datetime] = mapped_column(UtcDateTime, default=lambda: datetime.now(timezone.utc))