分享一个学习中的代码片段通过火山引擎Agent Plan的API Key langchain搭建一个简单的Agent助手代码如下fromlangchain_openaiimportChatOpenAIfromlangchain.agentsimportcreate_agent,AgentStatefromlangchain.agents.middlewareimportbefore_agent,after_agent,before_model,after_model,wrap_model_call,wrap_tool_callfromlanggraph.runtimeimportRuntimefromlangchain_core.toolsimporttoolfromdotenvimportload_dotenvimportos load_dotenv()llmChatOpenAI(modelark-code-latest,api_keyos.getenv(API_KEY),base_urlos.getenv(BASE_URL),)# TODO: Tools工具# tool是Python 的装饰器语法把普通函数自动转换成LangChain的BaseTool工具对象tool(description获取体重返回值是整数单位千克)defget_weight()-int:return90tool(description获取身高返回值是整数单位厘米)defget_height()-int:return180tool(description获取名字)defget_name()-str:return马牛逼tool(description获取兴趣爱好)defget_interest()-str:return唱、跳、rap、、music# TODO: Middleware 中间件一共有6种before_agentdeflog_before_agent(state:AgentState,runtime:Runtime)-None:print(f\n[before_agent]Starting agent with {len(state[messages])}messages\n)after_agentdeflog_after_agent(state:AgentState,runtime:Runtime)-None:print(f\n[after_agent]Agent completed with {len(state[messages])}messages\n)before_modeldeflog_before_model(state:AgentState,runtime:Runtime)-None:print(f\n[before_model]About to call model with {len(state[messages])}messages\n)after_modeldeflog_after_model(state:AgentState,runtime:Runtime)-None:print(\nafter_model\n,state[messages][-1].content)wrap_model_call# 模型调用时拦截defmodel_call_hook(request,handler):print(\n模型调用了\n)returnhandler(request)wrap_tool_call# 工具调用时拦截deftool_call_hook(request,handler):print(\n工具调用了\n)print(f\n工具执行{request.tool_call[name]}\n)print(f\n工具执行传入参数{request.tool_call[args]}\n)returnhandler(request)agentcreate_agent(modelllm,tools[get_weight,get_height,get_name,get_interest],system_prompt你是信息助手 ,middleware[log_before_agent,log_after_agent,log_before_model,log_after_model,model_call_hook,tool_call_hook])print(agent.invoke({messages:[{role:user,content:告诉我信息}]}))
分享一个学习中的代码片段通过火山引擎Agent Plan的API Key langchain搭建一个简单的Agent助手代码如下fromlangchain_openaiimportChatOpenAIfromlangchain.agentsimportcreate_agent,AgentStatefromlangchain.agents.middlewareimportbefore_agent,after_agent,before_model,after_model,wrap_model_call,wrap_tool_callfromlanggraph.runtimeimportRuntimefromlangchain_core.toolsimporttoolfromdotenvimportload_dotenvimportos load_dotenv()llmChatOpenAI(modelark-code-latest,api_keyos.getenv(API_KEY),base_urlos.getenv(BASE_URL),)# TODO: Tools工具# tool是Python 的装饰器语法把普通函数自动转换成LangChain的BaseTool工具对象tool(description获取体重返回值是整数单位千克)defget_weight()-int:return90tool(description获取身高返回值是整数单位厘米)defget_height()-int:return180tool(description获取名字)defget_name()-str:return马牛逼tool(description获取兴趣爱好)defget_interest()-str:return唱、跳、rap、、music# TODO: Middleware 中间件一共有6种before_agentdeflog_before_agent(state:AgentState,runtime:Runtime)-None:print(f\n[before_agent]Starting agent with {len(state[messages])}messages\n)after_agentdeflog_after_agent(state:AgentState,runtime:Runtime)-None:print(f\n[after_agent]Agent completed with {len(state[messages])}messages\n)before_modeldeflog_before_model(state:AgentState,runtime:Runtime)-None:print(f\n[before_model]About to call model with {len(state[messages])}messages\n)after_modeldeflog_after_model(state:AgentState,runtime:Runtime)-None:print(\nafter_model\n,state[messages][-1].content)wrap_model_call# 模型调用时拦截defmodel_call_hook(request,handler):print(\n模型调用了\n)returnhandler(request)wrap_tool_call# 工具调用时拦截deftool_call_hook(request,handler):print(\n工具调用了\n)print(f\n工具执行{request.tool_call[name]}\n)print(f\n工具执行传入参数{request.tool_call[args]}\n)returnhandler(request)agentcreate_agent(modelllm,tools[get_weight,get_height,get_name,get_interest],system_prompt你是信息助手 ,middleware[log_before_agent,log_after_agent,log_before_model,log_after_model,model_call_hook,tool_call_hook])print(agent.invoke({messages:[{role:user,content:告诉我信息}]}))
