Invalid API key for custom base URL

[vanloc1808]

Please read this first

• Have you read the docs?Agents SDK docs yes
• Have you searched for related issues? Others may have faced similar issues. yes

Describe the bug

I am about to create an AsyncOpenAI client, with base_url to point to my endpoint, and my endpoint does not require authentication. However, the error of invalid API key shows.

Debug information

• Agents SDK version: (e.g. v0.0.3)
• Python version (e.g. Python 3.12.0)

Repro steps

from openai import AsyncOpenAI
from agents import set_default_openai_client

set_default_openai_client(
    AsyncOpenAI(
        base_url=<my-base-url>, 
        api_key="sk-bjkacbsajkcbajcakbcsjaascbkjsabcajkcbakjascbkjsc",
    ),
)

My endpoint does not require API key. In other implementation, I can still use a dummy API key for interacting with this.
However, when I call Runner.run(), it shows the error of invalid API key.

Expected behavior

The client should work without checking the API key in this case.

[venvoyv]

#122 set_default_openai_client(external_client) below, which seems to be working, but I'm now encountering some other issues that I'm not yet aware of, and I need to figure them out on my own.

[overstarry]

set_tracing_disabled(True)

[rayleighzhongrui]

set_use_responses_by_default(False)
否则会默认使用OpenAI最新发布的response api，如果custom base URL不支持的话就会报错

[Howe829]

set_use_responses_by_default(False) 否则会默认使用OpenAI最新发布的response api，如果custom base URL不支持的话就会报错

You are right, but it seems you have to import "set_use_responses_by_default" from a protected file.

from agents.models._openai_shared import set_use_responses_by_default

[Howe829]

set_use_responses_by_default(False) 否则会默认使用OpenAI最新发布的response api，如果custom base URL不支持的话就会报错

You are right, but it seems you have to import "set_use_responses_by_default" from a protected file.

from agents.models._openai_shared import set_use_responses_by_default

After I read the source code, I think the correct answer is you have to set the OpenAI api type to "chat_completions" if your api doesn't support "responses"

from agents import set_default_openai_api
set_default_openai_api("chat_completions")

In addition, I think it's better to rename the func "set_default_openai_api" to "set_default_openai_api_type".

[penetest]

something is ok:

but it is also show some error.

[rayleighzhongrui]

set_use_responses_by_default(False) 否则会默认使用OpenAI最新发布的response api，如果custom base URL不支持的话就会报错

You are right, but it seems you have to import "set_use_responses_by_default" from a protected file.
from agents.models._openai_shared import set_use_responses_by_default

After I read the source code, I think the correct answer is you have to set the OpenAI api type to "chat_completions" if your api doesn't support "responses"

from agents import set_default_openai_api
set_default_openai_api("chat_completions")
In addition, I think it's better to rename the func "set_default_openai_api" to "set_default_openai_api_type".

是的，为了解决这个问题两个方法都可以

[rm-openai]

These docs should help: https://openai.github.io/openai-agents-python/models/#using-other-llm-providers

[vanloc1808]

#122 set_default_openai_client(external_client) below, which seems to be working, but I'm now encountering some other issues that I'm not yet aware of, and I need to figure them out on my own.

I tried set_default_openai_client before, but the problem still persisted

[vanloc1808]

These docs should help: https://openai.github.io/openai-agents-python/models/#using-other-llm-providers

Thanks

[puke3615]

Summary:

Core setting:

from agents import set_default_openai_api, set_tracing_disabled

set_default_openai_api("chat_completions")
set_tracing_disabled(True)

Full demo code:

import asyncio
from agents import Agent, Runner, function_tool, set_default_openai_api, set_tracing_disabled

set_default_openai_api("chat_completions")
set_tracing_disabled(True)

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



async def main():
    agent = Agent(
        name="AI Assistant",
        instructions="You are a helpful assistant",
        model="gpt-4o",
        tools=[get_weather],
    )

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

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

Output:

The weather in Tokyo is currently sunny.