How to expose as remote tool which can be called by remote Agent

[weiminw]

I use langserve develop a chain , and expose as remote tool. my friend wants to call my chain in his agent, how to do it?

Joke chain:

#!/usr/bin/env python
from fastapi import FastAPI
from langchain.prompts import ChatPromptTemplate
from langchain.chat_models import ChatAnthropic, ChatOpenAI
from langserve import add_routes
llm = ChatOpenAI(
    openai_api_base=f"http://192.168.1.201:18001/v1",
    openai_api_key="EMPTY",
    model="gpt-3.5-turbo",
    temperature=0.5,
    top_p="0.3",
    default_headers={"x-heliumos-appId": "general-inference"},
    tiktoken_model_name="gpt-3.5-turbo",
    verbose=True,
)

app = FastAPI(
  title="LangChain Server",
  version="1.0",
  description="A simple api server using Langchain's Runnable interfaces",
)


prompt = ChatPromptTemplate.from_template("tell me a joke about {topic}")
add_routes(
    app,
    prompt | llm,
    path="/joke",
)

if __name__ == "__main__":
    import uvicorn

    uvicorn.run(app, host="0.0.0.0", port=8000)

Agent:

from langchain.agents import initialize_agent, AgentType

from langchain_community.chat_models import ChatOpenAI
from langserve import RemoteRunnable

from langchain.tools import Tool

llm = ChatOpenAI(
    openai_api_base=f"http://xxxx:xxx/v1",
    openai_api_key="EMPTY",
    model="gpt-3.5-turbo",
    temperature=0.5,
    top_p="0.3",
    tiktoken_model_name="gpt-3.5-turbo",
    verbose=True,
)

remote_tool = RemoteRunnable("http://xxx:xxx/joke/")
tools = [
    Tool.from_function(
        func=remote_tool.invoke,
        name="joke",
        description="用户要求讲笑话的时候使用该工具",
        # coroutine= ... <- you can specify an async method if desired as well
    ),
]
agent = initialize_agent(
    tools, llm, agent=AgentType.ZERO_SHOT_REACT_DESCRIPTION, verbose=True
)

result = agent.run("讲一个关于坐出租车的笑话")
print(result)

agent always got error: because not there is not valid input to remote tool.

[eyurtsev]

The best way right now is to wrap it in another function. This allows you to populate the doc-string and specify the type signature appropriately.

from langchain.tools import tool

@tool
def tell_joke(topic: str) -> str:
   """A tool that can be used to tell a joke."""
   return remote_tool.invoke({'topic': topic})

tools = [
  tell_joke
] 
...

[weiminw]

great thanks to you