models permission required

[sowjanyagunupuru]

Hii,
I followed the steps outlined in Setting Up GitHub Models for Theia AI.

I used both OpenAI o4-mini and OpenAI GPT-4o, generating separate GitHub tokens to be specified under apiKey in the configuration.
{
"model": "o4-mini",
"url": "https://models.inference.ai.azure.com",
"id": "o4-mini",
"apiKey": "yourGithubToken",
"enableStreaming": true,
"developerMessageSettings": "developer"
}
In the AI Configuration [Alpha] view, I selected o4-mini as the LLM for the agent Coder.

Now, in the Theia AI Chat, after pinning the agent as Coder and providing a prompt, I received the following error message:

401 Unauthorized: The models permission is required to access this endpoint.

What’s confusing is that I followed the same exact process earlier using GPT-4o, and it worked perfectly when I first started exploring Theia AI Chat.

However, suddenly, I started receiving the above error.
Thinking the issue might be due to an expired GitHub token, I regenerated the token — but the error persisted.

Then I switched to using the o4-mini model with a new token, hoping it would resolve the issue. Unfortunately, the problem remains the same.

• Could you help identify why I'm now getting a 401 Unauthorized error requiring the models permission, even though the setup previously worked?

• I am also trying to understand how tool calls are handled within Theia AI (e.g., how the request/response structure works when using functions like writeFile).

It would be helpful to know which files in the Theia AI codebase are responsible for Handling tool call responses (like createToolCallResponseContent)

Understanding these would help me integrate or modify support for custom models to work properly with tool calls, just like the built-in models or the OpenAIModels

Thank you in advance for your help.

[planger]

Hi,

it looks like GitHub now requires fine-grained Personal Access Tokens (PATs) to have explicit models:read permissions in order to use their models. See announcement.

So even though your setup worked earlier (likely with a coarse-grained token or before the permission was enforced), newer tokens must include models:read to continue working.

1. Create a new fine-grained PAT via GitHub Developer Settings.
2. Select the models:read permission when setting the scopes.
3. Replace your old token with the new one in the AI configuration under apiKey.

Coarse-grained tokens (the older style) don’t require this permission and should still work.

We'll update the blog post, thanks for letting us know!

I hope this helps!

[sowjanyagunupuru]

Thank you for your response. The solution worked very well.