Support for AzureOpenAI

[akolk]

Can you check if the AZURE_OPENAI_ENDPOINT environment variable exists and then use the AzureOpenAI instead of OpenAI client? We have an AzureOpenAI account that we use a lot.

change:
self._client = openai.OpenAI(api_key=self.api_key)

to:
self._client = openai.OpenAI(api_key=self.api_key) if os.getenv("AZURE_OPENAI_ENDPOINT") is None else openai.AzureOpenAI(api_key=self.api_key)

[sbslee]

@akolk, thank you for the suggestion. I'm not familiar with the Azure OpenAI API, but I'm open to giving it a try. From a quick search, it looks like in addition to AZURE_OPENAI_ENDPOINT and AZURE_OPENAI_API_KEY, you also need OPENAI_API_VERSION. Can you confirm that these are the only parameters I need to consider?

[akolk]

The only thing that you need to check is the AZURE_OPENAI_ENDPOINT environment variable. The model name can be set already in the Chat() function. That should be the same name as the deployment on that endpoint, so that is something that you don't need to worry about. If some one uses AZURE_OPENAI they will know what to do.

But it would be nice to may be add a api_version parameter to the Chat() function, so one can override if needed.

[sbslee]

@akolk, I see. As requested, I've created the feature/azure-openai-support branch and added support for the Azure OpenAI API as you suggested. Could you please test it out and let me know if it works as expected?

[akolk]

Thanks, I will try it today and let you know if this works.

[akolk]

Did a quick check and the problem seems to be that the openai and azureopenai calls are fully compatible for assistants (code_interpreter, web_search etc) and containers. If I disable them with allow_*=False, it works. Currently checking what can be done to support that

[akolk]

So it seems that not everything that streamlit-openai relies on is available just yet in the azure openai implementation:
https://learn.microsoft.com/en-us/azure/ai-services/openai/how-to/responses?tabs=python-key#model-support

Not currently supported:

The web search tool
Fine-tuned models
Image generation via streaming. Coming soon.
Images can't be uploaded as a file and then referenced as input. Coming soon.
There's a known issue with performance when background mode is used with streaming. The issue is expected to be resolved soon.

[sbslee]

@akolk, thank you for your thorough investigation! Since many of OpenAI's core functionalities still need to be ported to the Azure OpenAI API, I’ll hold off on this until that happens.

[akolk]

This is what my code in chat.py looks like:

self._client = openai.OpenAI(api_key=self.api_key) if os.getenv("AZURE_OPENAI_ENDPOINT") is None else openai.OpenAI(api_key=os.getenv("AZURE_OPENAI_API_KEY"), base_url=os.getenv("AZURE_OPENAI_ENDPOINT")+ "openai/v1/", default_query={"api-version": "preview"})

The following settings are in app.py:
MODEL=gpt-4.1

allow_image_generation=False,
  model=os.getenv("MODEL", "o3"), # Select a reasoning model
  allow_web_search=False, # Disable web search
  allow_code_interpreter=False

I needed a demo for the MCP servers and that works, it works better if you specify a reasoning model but for now it is ok :-)