Getting openai.BadRequestError when trying to run with local vllm openai entrypoints

[amitli1]

I'm trying to run the get_weather example and to work with my LLM server:

import asyncio
import os
from agents import Agent, Runner, function_tool, set_tracing_disabled, OpenAIChatCompletionsModel
from openai import AsyncOpenAI

async def main():


    set_tracing_disabled(True)
    llm_url = "http://localhost:8090/v1"
    llm_model = "models/Qwen3-4B-AWQ"
    api_key = "EMPTY"
    os.environ["OPENAI_API_KEY"] = api_key

    client     = AsyncOpenAI(base_url= llm_url, api_key=api_key)
    chat_model = OpenAIChatCompletionsModel(model=llm_model, openai_client=client)


    @function_tool
    def get_weather(city: str) -> str:
        return f"The weather in {city} is sunny."


    agent = Agent(
        name="Hello world",
        model=chat_model,
        instructions="You are a helpful agent.",
        tools=[get_weather],
    )

    result = await Runner.run(agent, input="What's the weather in Tokyo?")
    print(result.final_output)


if __name__ == "__main__":
    asyncio.run(main())`

And I'm getting error:

raise self._make_status_error_from_response(err.response) from None
openai.BadRequestError: Error code: 400 - {'object': 'error', 'message': '"auto" tool choice requires --enable-auto-tool-choice and --tool-call-parser to be set', 'type': 'BadRequestError', 'param': None, 'code': 400}

How can I run this example with my llm ?

[ihower]

@amitli1 I think this is not related to the OpenAI Agents SDK. The reason is already explained in the error message from vllm.

Your vLLM server must be started with the following flags enabled: --enable-auto-tool-choice and --tool-call-parser

Please refer to the vLLM doc: https://docs.vllm.ai/en/stable/features/tool_calling/