采用向量数据库实现长期记忆

app/agent.py

@@ -6,16 +6,15 @@ AI Agent 服务类 - 支持多模型动态切换
 import os
 from dotenv import load_dotenv
 from langchain_community.chat_models import ChatZhipuAI
-from langchain_core.messages import HumanMessage
 from langchain_openai import ChatOpenAI
 from pydantic import SecretStr
+from langgraph.checkpoint.postgres.aio import AsyncPostgresSaver
 
 # 本地模块
 from app.graph_builder import GraphBuilder, GraphContext
 from app.tools import AVAILABLE_TOOLS, TOOLS_BY_NAME
 from app.logger import debug, info, warning, error
-from langgraph.checkpoint.postgres.aio import AsyncPostgresSaver
-from langgraph.store.postgres.aio import AsyncPostgresStore
+
 
 load_dotenv()
 
@@ -23,15 +22,13 @@ load_dotenv()
 class AIAgentService:
     """异步 AI Agent 服务，支持多模型动态切换，使用外部传入的 checkpointer"""
 
-    def __init__(self, checkpointer: AsyncPostgresSaver, store: AsyncPostgresStore):
+    def __init__(self, checkpointer: AsyncPostgresSaver):
         """
         初始化服务
         Args:
             checkpointer: 已经初始化的 AsyncPostgresSaver 实例
-            store: 已经初始化的 AsyncPostgresStore 实例
         """
         self.checkpointer = checkpointer
-        self.store = store
         self.graphs = {}  # 存储不同模型对应的 graph 实例
 
     def _create_zhipu_llm(self):
@@ -68,19 +65,19 @@ class AIAgentService:
         # vLLM 服务地址：优先从环境变量读取，适配 Docker、FRP 穿透和本地开发
         vllm_base_url = os.getenv(
             "VLLM_BASE_URL",
-            "http://115.190.121.151:18000/v1"
+            "http://localhost:8081/v1"
         )
         
         return ChatOpenAI(
             base_url=vllm_base_url,
-            api_key=SecretStr(os.getenv("VLLM_LOCAL_KEY", "")),
+            api_key=SecretStr(os.getenv("LLAMACPP_API_KEY", "token-abc123")),
             model="gemma-4-E2B-it",
             timeout=60.0,          # 请求超时时间（秒）
             max_retries=2,          # 失败后自动重试次数
         )
 
     async def initialize(self):
-        """预编译所有模型的 graph（使用传入的 checkpointer 和 store）"""
+        """预编译所有模型的 graph（使用传入的 checkpointer）"""
         model_configs = {
             "zhipu": self._create_zhipu_llm,
             "deepseek": self._create_deepseek_llm,
@@ -92,7 +89,7 @@ class AIAgentService:
                 info(f"🔄 正在初始化模型 '{model_name}'...")
                 llm = llm_creator()
                 builder = GraphBuilder(llm, AVAILABLE_TOOLS, TOOLS_BY_NAME).build()
-                graph = builder.compile(checkpointer=self.checkpointer, store=self.store)
+                graph = builder.compile(checkpointer=self.checkpointer)
                 self.graphs[model_name] = graph
                 info(f"✅ 模型 '{model_name}' 初始化成功")
             except Exception as e:
@@ -121,14 +118,27 @@ class AIAgentService:
                 "elapsed_time": float   # 调用耗时（秒）
             }
         """
+        # 尝试使用指定模型，如果不可用则循环尝试其他模型
         if model not in self.graphs:
-            fallback_model = next(iter(self.graphs.keys()))
-            warning(f"警告: 模型 '{model}' 不可用，已切换到 '{fallback_model}'")
-            model = fallback_model
+            warning(f"警告: 模型 '{model}' 不可用，尝试切换到其他可用模型")
+            found = False
+            for available_model in self.graphs.keys():
+                try:
+                    # 这里可以添加额外的模型可用性检查逻辑
+                    model = available_model
+                    found = True
+                    info(f"已切换到可用模型: '{model}'")
+                    break
+                except Exception as e:
+                    warning(f"模型 '{available_model}' 也不可用: {str(e)}")
+                    continue
+            
+            if not found:
+                raise RuntimeError(f"错误: 没有任何可用的模型。当前注册的模型: {list(self.graphs.keys())}")
 
         graph = self.graphs[model]
         config = {"configurable": {"thread_id": thread_id}}
-        input_state = {"messages": [HumanMessage(content=message)]}
+        input_state = {"messages": [{"role": "user", "content": message}]}
         context = GraphContext(user_id=user_id)
         
         result = await graph.ainvoke(input_state, config=config, context=context)