[Config Support]: OpenAI GenAI provider ignores base_url — all requests sent to api.openai.com

[joelmansford]

Describe the problem you are having

When configuring the genai section with provider: openai and a custom base_url pointing to a self-hosted OpenAI-compatible API (e.g. vLLM, Ollama), the base_url value is accepted by config validation and stored internally, but never passed to the OpenAI() client constructor. All GenAI requests are sent to api.openai.com instead of the configured endpoint.

Version

0.16.4

Frigate config file

genai:
     enabled: true
     provider: openai
     api_key: sk-any-key
     base_url: http://my-local-server:8002/v1
     model: my-local-model

Relevant Frigate log output

n/a

Relevant go2rtc log output

n/a

Frigate stats

No response

Operating system

Home Assistant OS

Install method

Home Assistant Add-on

docker-compose file or Docker CLI command

n/a

Object Detector

Coral

Screenshots of the Frigate UI's System metrics pages

No response

Any other information that may be helpful

In frigate/genai/openai.py, the _init_provider method only passes api_key to the OpenAI() constructor:

def _init_provider(self):
    return OpenAI(api_key=self.genai_config.api_key)

The base_url from self.genai_config.base_url is never included. The fix would be:

def _init_provider(self):
    return OpenAI(
        api_key=self.genai_config.api_key,
        base_url=self.genai_config.base_url,
    )

Workaround

The Python openai library automatically reads the OPENAI_BASE_URL environment variable when no explicit base_url is passed to the constructor. Frigate supports setting environment variables via the top-level environment_vars config section:

Additional Context

I understand that v0.17.x introduces provider_options.base_url which may address this. However, the base_url field in v0.16.x silently accepts the value without any warning that it won't be used, which makes debugging very difficult — config validation passes, the add-on starts cleanly, and the only clue is 401 errors buried in logs pointing at api.openai.com.

If this won't be backported to 0.16.x, it would be helpful to either:

1. Log a warning when base_url is set with the openai provider (explaining it's not used), or
2. Document the environment_vars workaround in the GenAI docs.

[inkeep-ai-assistant[bot]]

You've identified a known issue with the OpenAI provider in Frigate 0.16.x. The base_url configuration option is not passed to the OpenAI client constructor, causing all requests to go to api.openai.com instead of your custom endpoint(1).

Workaround

The documented workaround is to set the OPENAI_BASE_URL environment variable(2)(3). You can do this using Frigate's environment_vars configuration section(4):

environment_vars:
  OPENAI_BASE_URL: http://my-local-server:8002/v1

genai:
  enabled: true
  provider: openai
  api_key: sk-any-key
  model: my-local-model

(4)

The Python openai library will automatically use this environment variable when no explicit base_url is provided to the constructor(3).

Additional Notes

This issue has been reported in the context of OpenRouter and other OpenAI-compatible APIs(1). A community member confirmed that setting OPENAI_BASE_URL as an environment variable works as a solution(3).

For Home Assistant add-on users specifically, there was discussion about adding the base_url directly to the add-on configuration to simplify this setup(1).

📚 Sources:

1. [Beta Support]: GenAI OpenAI Provider ignores OpenRouter base URL and calls api.openai.com
2. Generative AI
3. genai seems to work best with openai provider and not so well with the ollama provider
4. Frigate Configuration

---

Have another question? Just tag @inkeep.

[hawkeye217]

The AI's response here is correct. You need to use OPENAI_BASE_URL. This is noted in the official 0.16.x docs here: https://docs.frigate.video/configuration/genai#configuration-3