All contributors have signed the CLA ✍️ ✅
_{Posted by the CLA Assistant Lite bot.}

I have read the CLA Document and I hereby sign the CLA

@codex review

For the record, my main apprehension with this change is that while the built-in definition for ollama is quite simple and straightforward to override:

The built-in definition for the openai model provider is much more complex:

For ollama, it seems that the main thing one needs to override is the base_url, which can also be controlled by the CODEX_OSS_BASE_URL and CODEX_OSS_PORT environment variables.

Similarly, we currently honor the OPENAI_BASE_URL environment variable so you can modify that one field of the openai model provider without accidentally messing up the other fields.

That said, I don't love the environment variables.

Would it be more attractive to add support for openai_base_url and ollama_base_url config options instead?

/cc @etraut-openai who I think was also hoping to eliminate support for OPENAI_BASE_URL

@bolinfest your concern about openai complexity makes sense. For the ollama case in #8240, users just need to point at a remote server - they shouldn't have to know about all the internal fields.

Thoughts:

• The current fix matches the documented behavior ("override or extend"), but I agree it's a footgun for openai
• ollama_base_url would solve Ollama support assumes localhost, ignoring config.toml #8240 cleanly, though it's more narrow - wouldn't help if someone wanted to also override wire_api for example
• A middle ground could be merging user fields into built-in defaults (so [model_providers.ollama] with just base_url keeps the other defaults), but that's more complex

Happy to go whichever direction you prefer. The *_base_url config options seem like the simplest path if the goal is just enabling remote servers without exposing the full provider complexty.