OpenAI Agents SDK

Use the Agents SDK Responses API adapter with an AsyncOpenAI client that targets the local gateway:

The executable compatibility suite uses openai-agents 0.22.2.

import asyncio

from agents import Agent, RunConfig, Runner
from agents.models.openai_responses import OpenAIResponsesModel
from openai import AsyncOpenAI


async def main() -> None:
    async with AsyncOpenAI(
        base_url="http://127.0.0.1:8080/v1",
        api_key="local",
        max_retries=0,
    ) as client:
        model = OpenAIResponsesModel(
            model="Qwen/Qwen2.5-0.5B-Instruct",
            openai_client=client,
        )
        agent = Agent(name="PLLM agent", instructions="Reply briefly.", model=model)
        result = await Runner.run(
            agent,
            "Describe the trusted client boundary.",
            run_config=RunConfig(tracing_disabled=True),
        )
    print(result.final_output)


asyncio.run(main())

This path passes a repository integration test against a real local Responses API gateway. The gateway accepts text and local function-tool traffic. Do not enable hosted tools or assume support for every Agents SDK model feature. Tracing is disabled because this configuration does not use an OpenAI-hosted tracing endpoint.