Getting 401 request error with open ai service account keys

[peppermint-ai-lab]

Initial Checks

• I confirm that I'm using the latest version of Pydantic AI

Description

We recently created a service account key with open ai and started using it instead of user account key. However, after few minutes (~15 minutes) we see 401 errors

openai.AuthenticationError: Error code: 401 - {'error': {'message': 'Incorrect API key provided:  You can find your API key at https://platform.openai.com/account/api-keys.', 'type': 'invalid_request_error', 'param': None, 'code': 'invalid_api_key'}}

We had deployed to production (cloud run in GCP) and had validated that the deployment was funcitoning. During smoke testing, after few minutes, the error occurred. We redeployed and observed similar pattern.

I am wondering if this is an issue similar to #1186

Example Code

Python, Pydantic AI & LLM client version

python - 3.12
Pydantic and llm client - 0.0.43

[Kludex]

That doesn't seem on our side. Please try to bump OpenAI package to see if it solves the problem.

If it doesn't, please provide an. MRE.

[peppermint-ai-lab]

Some more info

• I have two packages pydantic-ai-slim[openai]==0.0.43, pydantic-ai-slim[vertexai]==0.0.43
• the error occurs when second agent is activated. In our app, agents are activated serially after the previous one completes

I did some further debugging. Here are my findings

1. Locally when I use a .env file it seems to be working fine
2. With docker image locally it seems to work fine when passing env as params
3. On google cloud with cloud run,

• If I use env variable or secret with general Agent constructor it fails on the second agent
• If I use env variable or hardcoded api key but by instantiating provider, it works perfectly fine

Another thing to notice is that the error indicates the start of api key, but that is not the start of my api key. I did print out the api key before agent initing and it looked fine.

Incorrect API key provided: ya29.c.c*

Let me know if I can provide further info.

What is MRE?

[Kludex]

Well, in your description above you said the issue was with OpenAI, not with Vertex...

MRE is a minimal reproducible example, that I can just copy the code and run on my machine.

[peppermint-ai-lab]

Here you go (https://github.com/peppermint-ai-lab/pydantic-1221-mre)

As to what is happening, my guess is running agent with vertex-ai is resetting the Open AI API key. I don't know if this issue happens with other models as well.

[Kludex]

I've fixed this on #1242.

Thanks for the help!