Add support for GitHub Models endpoint validation

[Gerson-Cabbage]

Linked Issue

N/A

Type of Change

• New feature
• Bug fix
• Documentation update
• Refactor

Summary

Add support for GitHub Models as a built-in provider.

Changes

• Added automatic detection of GitHub Models endpoints (models.github.ai).
• Added a list of supported GitHub Models.
• Allows users to connect using a GitHub Personal Access Token.

Supported Models

• GPT-4o
• GPT-4o-mini
• o1
• o1-mini
• o3-mini
• Meta Llama models
• Microsoft Phi models
• Mistral models
• DeepSeek models
• Cohere models

Checklist

• I searched existing issues and pull requests for duplicates.
• I tested the changes locally.
• The feature works as expected.

How to Test

1. Open Odysseus.

2. Add a new API endpoint.

3. Use:

   Base URL:
   https://models.github.ai/inference

4. Use a valid GitHub Personal Access Token.

5. Verify that the provider is detected as github.

6. Verify that the supported models appear in the model list.

7. Select a model and confirm successful inference requests.

[github-actions]

⚠️ PR description — action needed

The following required sections are missing or incomplete. Please update the PR description to address them:

• Linked Issue — add a reference like Fixes #NNN, a bare #NNN, or a link to the issue.
• Type of Change — check at least one box.
• Checklist — check the duplicate-search box to confirm you searched existing issues and PRs.

---

This comment is deleted automatically once all sections are complete.

[RaresKeY]

I think we should detect model capabilities and vendors/providers more deterministically instead of continuing to rely on model-name matching. Some providers may not expose capability metadata at all, and others may expose it in different native shapes, so Odysseus needs a normalized capability layer rather than more one-off heuristics. This relates to #2739.

[RaresKeY]

Pre-review automated code review

Findings

[high] routes/model_routes.py:267 - GitHub Models exposes model IDs that the inference API will reject

Why it matters: The fallback list returns bare IDs such as gpt-4o, o1, and DeepSeek-R1. Odysseus sends the selected value unchanged as the request model, but GitHub Models expects publisher-qualified catalog IDs such as openai/gpt-4.1. That can make models appear selectable while inference fails for the selected model.

Suggested fix: Fetch https://models.github.ai/catalog/models and cache the returned id values, or make the curated fallback use the same publisher-qualified IDs. Add a regression test that _probe_endpoint("https://models.github.ai/inference", ...) returns sendable GitHub model IDs.

Validation

Ran: gh pr view, gh pr checks, gh pr diff --patch, GitHub issue/review comment API reads, exact PR-head file reads, coordinator PR-diff verification, and official GitHub Models docs checks.

Not run: pytest, app/browser verification, or live inference with a GitHub PAT. Direct unauthenticated catalog probe was rate-limited and the inference models probe returned 401.

PR hygiene

PR target/template status: targets dev; PR description check is failing and the PR has needs work. Existing comments include bot template feedback and a maintainer note about deterministic provider/capability detection.

Note

Automated pre-review test; I manually checked this before posting.