Fix an OpenAI Agents SDK 404 at /responses.

The OpenAI Agents SDK uses the Responses API by default. Setting OPENAI_BASE_URL changes the client endpoint, but does not select Chat Completions. If a compatible gateway implements only POST /v1/chat/completions, an agent run can send POST /v1/responses and get a 404.

Check the actual request path

Read the method and final URL from the error or a local HTTP capture before changing a key:

A 404 alone does not prove an API-family mismatch. A wrong host, region, deployment path or API version can also return 404.

Select Chat Completions for a compatible endpoint

First set the provider’s documented base URL, key and exact model ID in the environment of the process that runs the agent:

export OPENAI_BASE_URL="https://api.example.com/v1"
export OPENAI_API_KEY="YOUR_PROVIDER_KEY"
export MODEL_ID="MODEL_ID_FROM_PROVIDER"

For one global client configured from OPENAI_API_KEY and OPENAI_BASE_URL, set the API family explicitly:

import os
from agents import Agent, Runner, set_default_openai_api

set_default_openai_api("chat_completions")
agent = Agent(name="Assistant", model=os.environ["MODEL_ID"])
result = await Runner.run(agent, "Reply with one short sentence.")
print(result.final_output)

Choose Chat Completions for one agent

Use an explicit client and OpenAIChatCompletionsModel when different agents need different endpoints. The SDK docs recommend disabling tracing when the configured provider does not supply an OpenAI tracing key.

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

set_tracing_disabled(True)
client = AsyncOpenAI(
    base_url=os.environ["OPENAI_BASE_URL"],
    api_key=os.environ["OPENAI_API_KEY"],
)
model = OpenAIChatCompletionsModel(
    model=os.environ["MODEL_ID"], openai_client=client
)
agent = Agent(name="Assistant", model=model)

async def main():
    result = await Runner.run(agent, "Reply with one short sentence.")
    print(result.final_output)

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

Fail fast on Responses-only features

When using the built-in OpenAI provider, OpenAIProvider(use_responses=False, strict_feature_validation=True) makes the API choice explicit and catches several Responses-only inputs instead of silently dropping them. With MultiProvider, the corresponding option is openai_strict_feature_validation=True.

from agents import Agent, OpenAIProvider, RunConfig, Runner

provider = OpenAIProvider(
    use_responses=False,
    strict_feature_validation=True,
)
agent = Agent(name="Assistant", model="MODEL_ID_FROM_PROVIDER")
result = await Runner.run(agent, "Reply with one short sentence.",
    run_config=RunConfig(model_provider=provider))

Responses-only fields include previous_response_id, conversation_id, the Responses prompt field and non-text-only tool outputs. Built-in Responses tools such as hosted web or file search do not become available through Chat Completions. Standard function tools are a separate feature and still depend on the gateway and model.

Separate tracing errors from model API errors

If the model provider is non-OpenAI and the SDK reports a 401 while exporting traces, that request goes to the tracing service; it is separate from the model call. Disable tracing or configure an appropriate tracing processor. Do not put a third-party provider key into OpenAI tracing settings.

ChinaGenAI endpoint

ChinaGenAI documents POST /v1/chat/completions, not /v1/responses. Use the Chat Completions setting above and check the documented endpoint limits and feature compatibility before connecting an app.

Check the SDK documentation for your installed version

• OpenAI Agents SDK models: Responses default, Chat Completions options and limitations
• OpenAI Agents SDK configuration: custom client and API selection
• OpenAIProvider reference: use_responses and strict_feature_validation
• OpenAI API: compare Responses and Chat Completions paths and payloads

Official SDK documentation checked 1 November 2026. The examples are configuration guidance; no API key, provider account or model request was used.

Free API access and limits

Check the site free API access and limits before trying to obtain a site key. Read the current status and request limits before using a site key. It is not a source of official provider keys.

ChinaGenAI is not affiliated with the providers named here. This release contains setup guidance and explicitly labeled fictional examples, not claimed live model benchmarks.