feat:增加存储用户对话记录

agent.py

@@ -5,16 +5,24 @@ import sqlite3
 from langchain_tavily import TavilySearch
 from langgraph.checkpoint.sqlite import SqliteSaver
 
+from services.db_service import init_db
+
+from dotenv import load_dotenv
+import os
+
+load_dotenv()
+
 # 连接sqlite
-connection = sqlite3.connect("chief.db", check_same_thread=False)
+connection = sqlite3.connect(os.getenv("SQLITE_DB_PATH"), check_same_thread=False)
 # 初始化checkpointer
 checkpointer = SqliteSaver(connection)
 # 自动建表
 checkpointer.setup()
+init_db()
 
 # web搜索工具，使用tavily作为web搜索工具
 web_search = TavilySearch(
-    tavily_api_key="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",
+    tavily_api_key=os.getenv("TAVILY_API_KEY"),
     max_results=5,
     topic="general"
 )
@@ -30,10 +38,10 @@ system_prompt = """
     """
 
 model = init_chat_model(
-    model="qwen3-max-2026-01-23",
+    model=os.getenv("MODEL_NAME"),
     model_provider="openai",
-    base_url="https://dashscope.aliyuncs.com/compatible-mode/v1",
-    api_key="sk-52bcd98e9c1d45908437c4e8706eefff",
+    base_url=os.getenv("MODEL_BASE_URL"),
+    api_key=os.getenv("MODEL_API_KEY"),
     temperature=1.5,
 )
 
@@ -43,3 +51,13 @@ agent = create_agent(
     checkpointer=checkpointer,  # 记忆
     system_prompt=system_prompt  # 系统提示词
 )
+
+title_prompt = """
+请根据以下对话内容，生成一句不超过 20 字的标题，用于记录本次对话主题。
+只返回标题，不要解释。
+"""
+
+title_agent = create_agent(
+    model=model,
+    system_prompt=title_prompt
+)

model.py

@@ -1,4 +1,6 @@
 from pydantic import BaseModel
 
 class ChatRequest(BaseModel):
+    username: str
     message: str
+    thread_id: str

requirements.txt

@@ -5,5 +5,6 @@ langchain-openai~=1.2.2
 langchain-tavily~=0.2.18
 langgraph~=1.2.2
 langgraph-checkpoint-sqlite~=3.1.0
-pydantic~=2.12.5
+pydantic~=2.13.4
 langchain-core~=1.4.0
+python-dotenv~=1.2.2

router.py

@@ -1,10 +1,8 @@
-import uuid
-
 from fastapi import APIRouter
 from fastapi.responses import StreamingResponse
 
 from model import ChatRequest
-from service import get_messages, clear_messages, search_recipes
+from services.agent_service import get_messages, clear_messages, search_recipes
 
 router = APIRouter()
 
@@ -13,7 +11,7 @@ router = APIRouter()
 async def chat_endpoint(request: ChatRequest):
     """流式对话"""
     return StreamingResponse(
-        search_recipes(request.message, str(uuid.uuid4())),
+        search_recipes(request),
         media_type="text/event-stream"
     )
 

services/agent_service.py

@@ -1,23 +1,38 @@
 from langchain_core.messages import HumanMessage, AIMessageChunk, AIMessage
 
-from agent import agent, checkpointer
+from agent import agent, checkpointer, title_agent
+from model import ChatRequest
+from services.db_service import exists_session, save_session
+from utils import dicts_to_messages
 
 
-async def search_recipes(prompt: str, thread_id: str):
+def generate_title(messages: list[dict[str, str]]) -> str:
+    response = title_agent.invoke({
+        "messages": dicts_to_messages(messages)
+    })
+    return response["messages"][-1].content.strip()
+
+
+async def search_recipes(request: ChatRequest):
     """调用agent搜索食谱"""
-    print(f"[用户]: {prompt}, thread_id: {thread_id}")
+    print(f"[用户] {request.username}: {request.message}, thread_id: {request.thread_id}")
     try:
-        message = HumanMessage(content=prompt)
+        message = HumanMessage(content=request.message)
 
         # 流式调用Agent
         for chunk, metadata in agent.stream(
                 {"messages": [message]},
-                {"configurable": {"thread_id": thread_id}},
+                {"configurable": {"thread_id": request.thread_id}},
                 stream_mode="messages"
         ):
             if isinstance(chunk, AIMessageChunk) and chunk.content:
                 yield chunk.content
 
+        # 总结对话标题并保存
+        if not exists_session(request.thread_id, request.username):
+            title = generate_title(get_messages(request.thread_id))
+            save_session(request.thread_id, request.username, title)
+
     except Exception as e:
         print(f"\n[错误]: {str(e)}")
         yield "信息检索失败，试试看手动输入食物列表？"
@@ -60,6 +75,6 @@ def get_messages(thread_id: str) -> list[dict[str, str]]:
         if isinstance(msg, HumanMessage):
             result.append({"role": "user", "content": msg.content})
         elif isinstance(msg, AIMessage):
-            result.append({"role": "assistant", "content": msg.content})
+            result.append({"role": "ai", "content": msg.content})
 
     return result

services/db_service.py

@@ -0,0 +1,54 @@
+import sqlite3
+import os
+
+
+def init_db():
+    conn = sqlite3.connect(os.getenv("SQLITE_DB_PATH"))
+
+    # 业务表：会话
+    conn.execute("""
+    CREATE TABLE IF NOT EXISTS session (
+        id INTEGER PRIMARY KEY AUTOINCREMENT,
+        thread_id TEXT NOT NULL UNIQUE,
+        username TEXT NOT NULL,
+        title TEXT NOT NULL,
+        created_time DATETIME DEFAULT CURRENT_TIMESTAMP
+    );
+    """)
+
+    conn.commit()
+    conn.close()
+
+
+def exists_session(thread_id: str, username: str) -> bool:
+    conn = sqlite3.connect(os.getenv("SQLITE_DB_PATH"))
+    cursor = conn.cursor()
+
+    cursor.execute(
+        """
+        SELECT 1
+        FROM session
+        WHERE thread_id = ? AND username = ?
+        LIMIT 1
+        """,
+        (thread_id, username)
+    )
+
+    result = cursor.fetchone()
+    conn.close()
+
+    return result is not None
+
+
+def save_session(thread_id: str, username: str, title: str):
+    conn = sqlite3.connect(os.getenv("SQLITE_DB_PATH"))
+    conn.execute(
+        """
+        INSERT OR REPLACE INTO session
+        (thread_id, username, title)
+        VALUES (?, ?, ?)
+        """,
+        (thread_id, username, title)
+    )
+    conn.commit()
+    conn.close()

test.py

@@ -0,0 +1,5 @@
+from agent import model
+
+if __name__ == '__main__':
+    response = model.invoke("西红柿、鸡蛋")
+    print(response)

utils.py

@@ -0,0 +1,11 @@
+from langchain_core.messages import HumanMessage, AIMessage
+
+
+def dicts_to_messages(messages: list[dict]):
+    result = []
+    for m in messages:
+        if m["role"] == "user":
+            result.append(HumanMessage(content=m["content"]))
+        elif m["role"] == "ai":
+            result.append(AIMessage(content=m["content"]))
+    return result