GitHub Copilot: Azure language model provider not showing API key configuration UI

[leewenjie]

Title: GitHub Copilot: Azure language model provider not showing API key configuration UI

Does this issue occur when all extensions are disabled?: No (requires GitHub Copilot extension)

• VS Code Version: 1.104.0-insider (user setup)
• OS Version: Ubuntu 24.04.2 LTS

Steps to Reproduce:

1. Install GitHub Copilot extension
2. Open VS Code Insiders
3. Open Command Palette (Ctrl+Shift+P)
4. Run "GitHub Copilot: Manage Language Models" command
5. Select "Azure" from the provider dropdown list
6. Observe that no API key configuration UI appears

Expected Behavior:

• Azure provider should display a configuration interface to enter Azure OpenAI endpoint and API key
• Should be able to configure personal Azure OpenAI resources for use with GitHub Copilot
• Should work similar to other providers like OpenAI, Anthropic that show proper configuration UIs

Actual Behavior:

• No configuration UI appears when selecting Azure provider in Copilot's language model management
• Cannot enter API key or endpoint information for personal Azure OpenAI resources
• Other providers (OpenAI, Anthropic, etc.) work correctly and show proper configuration interfaces
• Unable to use personal Azure OpenAI subscriptions with GitHub Copilot

Additional Information:

• This specifically affects GitHub Copilot's ability to use custom Azure OpenAI endpoints
• The Azure option appears in Copilot's language model provider list but lacks configuration capability
• No error messages are displayed
• Other language model providers in Copilot work as expected with proper configuration UIs
• This prevents users from using their own Azure OpenAI resources with GitHub Copilot

Extension Information:

• GitHub Copilot extension is installed and active
• Issue is specific to Copilot's language model management feature

Environment Details:

• Electron: 37.2.3
• Node.js: 22.17.0
• Chromium: 138.0.7204.100
• OS: Ubuntu 24.04.2 LTS (Linux x64)

[lramos15]

Add the following setting for the Azure model you want

  "github.copilot.chat.azureModels": {
    "<modelId>": {
      "name": "Model Name",
      "maxInputTokens": 64000,
      "maxOutputTokens": 16000,
      "toolCalling": true,
      "url": "<Model URL>",
      "vision": true
    }
  },

We're working to bring back the UI flow to make this new way of doing things more discoverable

[leewenjie]

Thanks for the suggestion.

[DanTup]

Logan Ramos (@lramos15) what is the "Model URL" supposed to look like? I tried this:

https://<MY SERVICE>.services.ai.azure.com/models/chat/completions?api-version=2024-05-01-preview

However I just get this error:

Request Failed: 400 {"error":{"code":"UnsupportedToolUse","message":"Request included unsupported tool use. Using tool is not supported by this model","details":"Request included unsupported tool use. Using tool is not supported by this model"}}

The model is gpt-oss-120b which I understand is supposed to support tool calling?

I can't find any other URL in the Azure Foundry control panel, and the Copilot help seems to just keep making up URLs that also don't work. In the console for any other URL I see "Unrecognized Azure deployment URL", which I guess is coming from:

https://github.com/microsoft/vscode-copilot-chat/blob/817a58890536e84af7de00990deb8dcc74dc2723/src/extension/byok/vscode-node/azureProvider.ts#L33=L39

[davidglassborow]

I used the Azure OpenAI endpoint Url and it worked fine

[lramos15]

Danny Tuppeny (@DanTup) Your endpoint looks correct. I cannot really tell you why it would be throwing that error if you put tool calling as false in the setting I assume it works. That makes me think it is a problem with the Azure version of GPT oss 120b not supporting tools

[davidglassborow]

Logan Ramos (@lramos15) My setup works fine for chat, but looks like custom models isn't supported for Agent mode, is that correct ?

[lramos15]

David Glassborow (@davidglassborow) It should work in agent mode if tool calling is enabled on your endpoint and in the settings

[anirugu]

By following this post, I am able to make non-openai model working but I see a problem with gpt-5-mini,

Sorry, your request failed. Please try again. Request id: xxxxx

Reason: Request Failed: 400 { "error": { "message": "Unsupported value: 'temperature' does not support 0.1 with this model. Only the default (1) value is supported.", "type": "invalid_request_error", "param": "temperature", "code": "unsupported_value" } }

I see similar issue is also mentioned here microsoft/vscode-copilot-release#7302

[lramos15]

Anirudha Gupta (@anirugu) You need to add thinking: true to the config in your settings

[lramos15]

This has been fixed. There's now UI shown