feat: add GPT-OSS model family support - #354
Conversation
- Add 'gpt-oss' to getModelFamily() in constants.js (backend) - Add 'gpt-oss' to isSupportedModel() in model-api.js (backend) - Add 'gpt-oss' to getModelFamily() in data-store.js (frontend) - Add GPT filter button in models.html (WebUI) - Add GPT-OSS group in model-dropdown.js (WebUI) Enables proxy to recognize and route GPT-OSS models (e.g. gpt-oss-120b-medium) served via Antigravity, alongside existing Claude and Gemini families.
|
Hi! 👋 Just checking in on this PR. All CI checks have passed and there are no merge conflicts — looks ready to go. This feature would unlock GPT-OSS model routing through the proxy, which is a nice addition alongside Claude and Gemini support. Would love to get this reviewed and merged when someone has a chance. Thanks for the great work! 🚀 |
|
FYI — I've opened PR #357 (fix: include tool call id for non-Claude, non-Gemini models) which is a companion fix needed for GPT-OSS (and any future non-Gemini model family) to work with tool calls. The issue: The fix changes |
Summary
Adds support for GPT-OSS models (e.g.
gpt-oss-120b-medium) served via the Antigravity Cloud Code API.Previously, only Claude and Gemini families were recognized — any other model family was silently filtered out as
'unknown'and hidden from the WebUI. This patch adds'gpt-oss'as a third recognized family.Changes
getModelFamily()returns'gpt-oss'when model ID containsgpt-ossisSupportedModel()acceptsgpt-ossfamily (enables/v1/modelslisting + quota tracking)getModelFamily()recognizesgpt-oss(no longer defaults to'other'which means hidden)GPT-OSSgroup in model selector dropdownMotivation
Google has begun serving GPT-OSS models (OpenAI open-source) through the Antigravity Cloud Code API alongside Claude and Gemini. Users with Antigravity accounts can now access
gpt-oss-120b-mediumand similar models, but the proxy rejects them with "Invalid model" because the family filter only recognizesclaudeandgemini.Testing
gpt-oss-120b-mediumappears inGET /v1/modelsresponsePOST /v1/messageswithmodel: gpt-oss-120b-mediumreturns 200 OK