Cannot use Azure OpenAI as a Chatcraft LLM provider - `<redacted>-eastus2.cognitiveservices.azure.com/models is not a valid URL. `

[yurhasko]

I've configured an Azure OpenAI service deployment in Azure AI Foundry and tried hitting the following request:

curl -X POST "https://<redacted>-eastus2.cognitiveservices.azure.com/openai/deployments/gpt-4.1/chat/completions?api-version=2025-01-01-preview" \
    -H "Content-Type: application/json" \
    -H "Authorization: Bearer $AZURE_API_KEY" \
    -d '{
        "messages": [
            {
                "role": "user",
                "content": "Test request to see whether this works"
            }
        ],
        "max_completion_tokens": 800,
        "temperature": 1,
        "top_p": 1,
        "frequency_penalty": 0,
        "presence_penalty": 0,
        "model": "gpt-4.1"
    }' | jq .

And it returns a valid result. However, when I try adding Azure OpenAI as a custom provider in Chatcraft, it says:

Looking at https://learn.microsoft.com/en-us/azure/ai-services/openai/reference#authentication, it seems that /models isn't available, and instead a per-model deployment URL approach is being used. Does it mean that Chatcraft doesn't have Azure OpenAI service support atm, or am I doing something wrong?

[Amnish04]

@yurhasko As far as I know, ChatCraft currently only supports OpenAI compatible providers, which is why the easiest door to most custom providers is via openrouter like https://openrouter.ai/provider/azure. But I doubt openrouter is flexible enough for your purpose.

One solution I can think of is setting up a serverless function that mimics openai style /models and /chat/completions endpoints over Azure API and use that as a custom provider. I see that Vercel's AI SDK makes it fairly easy to the point you just need to convert its results to openai style.
https://ai-sdk.dev/providers/ai-sdk-providers/azure

@humphd might know an even easier way, as I may have missed some ChatCraft developments over time. I think he's doing something similar in #743.

[tarasglek]

right re models endpoint. we could probably add a feature to hardcode model list into settings