Add native OpenAI API integration

[albamoleroo]

Hi,

I’d like to add native support for invoking OpenAI’s REST API directly within pywhyllm, enabling:

• Easier downstream integrations (multi-repo usage, orchestration tools, internal platforms)
• Greater control over request/response handling (streaming, tool/function calling, response format validation)
• A clearer abstraction boundary for adding additional model providers later (Anthropic, Azure OpenAI, Ollama, local backends, etc.)

If no one is already working on this, I’m happy to open a PR implementing an initial provider. Please let me know if someone is already working on this (I didn’t want to duplicate effort).

[emrekiciman]

Hi @albamoleroo Thanks very much for this.

Right now, we use the guidance library as the abstraction boundary to different model providers (See for example )

The configuration is relatively simple though, and I'm sure we could improve how we use this library and/or manage the abstraction layer.

What do you have in mind?

[albamoleroo]

Hi @emrekiciman

Thanks for clarifying! I understand that guidance serves as the abstraction layer for multiple providers. My main motivation for proposing a native OpenAI provider is that guidance currently has limited access to some OpenAI API features, such as detailed token-level logprobs.

For example, if you need full token-level logprobs, streaming responses, function calling, or JSON-mode outputs, these aren’t fully exposed through guidance. Using the OpenAI client directly would allow more granular control over these parameters.

Although I realize this is a specific use case, but I think having a native OpenAI provider could be generally useful for anyone working with OpenAI in workflows where guidance’s abstraction is too limiting. But if not let me know! :)

[emrekiciman]

Hi @albamoleroo , Ok, yes, that makes a lot of sense and sounds very useful. We'd welcome this, and would really appreciate a PR. Let me know if you have any questions about the current code as you implement this. Thank you!