[FEAT] - Support Azure OpenAI in addition to OpenAI

[tfenster]

What you'd like to see

I would like to use Azure OpenAI instead of pure OpenAI

Why you'd like to see it

That would allow me to use cagent in more contexts as Azure OpenAI has some of the bells and whistles expected in a corporate context

Workarounds?

no

Additional context

If I try to call my Azure OpenAI endpoint using the existing openai provider, I would do something like this

models:
  cloud-gpt-4o:
    provider: openai
    model: gpt-4o
    base_url: https://openai-testtf.openai.azure.com/openai/deployments/gpt-4o

But this causes an error 404

The reason most likely is that Azure OpenAI expects a version query parameter. E.g. a POST to https://openai-testtf.openai.azure.com/openai/deployments/gpt-4o/chat/completions?api-version=2025-01-01-preview works, whereas a POST to https://openai-testtf.openai.azure.com/openai/deployments/gpt-4o/chat/completions also causes a 404.

The OpenAI client library used in cagent (if I understand the code correctly) seems to also support Azure OpenAI, but with different methods, see e.g. https://github.com/sashabaranov/go-openai/blob/8e5611cc5efdc2533b80e5667b69741c3fad875c/README.md?plain=1#L617-L618

[dgageot]

cc @krissetto

[krissetto]

We'll get support for this merged in the next days, we just have to decide how to best define the azure specific details in the model spec. I opened a proposal PR in #117

[tfenster]

Just to confirm, the proposed solution in the PR works for me both on Linux (WSL) and Windows with this config:

version: "1"

agents:
  root:
    description: An agent that talks like a pirate
    instruction: Always answer by talking like a pirate.
    model: cloud-gpt-4o

models:
  cloud-gpt-4o:
    provider: openai
    model: gpt-4o
    base_url: https://openai-testtf.openai.azure.com/
    provider_opts:
      azure_api_version: 2025-01-01-preview