feat(git): add model selection for commit message generation#451
Conversation
3a8bedc to
6d41bb0
Compare
|
YES PLS!! |
|
Why do we want that? It's adding complexity for a magic feature |
I've found the feature itself to be helpful but that using the same model I was using for feature work to be slow (sometimes it takes 7-8s) + felt like overkill for something so small that doesn't involve any real complexity. My assumption is that using a smaller model in this case also saves on usage for users on a budget. I had considered rewiring it to just pick a smaller model but I feel like that would potentially be more complex/error prone. Maybe it would be cleaner if this was tucked away in Settings instead of at the usage site? |
|
@joshhbk Yeah I think it should be a settings in the git section probably |
Allow users to choose which model generates commit messages via a dropdown in the git diff panel. The selection persists in app settings and falls back to the workspace default when the chosen model is no longer available.
4d24982 to
bd82a9d
Compare
Move the commit-message model dropdown from the git diff panel to Settings > Git. Source the model list from the existing codex-section fetch to avoid duplicate getModelList fan-out. Replace the old normalize-and-persist hook with a pure runtime guard (effectiveCommitMessageModelId) that validates the saved preference against the active workspace's models without overwriting the persisted setting — so switching workspaces no longer silently clears the choice.
bd82a9d to
8bd315f
Compare
|
@Dimillian updated! |
|
@codex review |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 3572101173
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Summary
Currently commit messages are generated with whatever the default model is. Picking one of the cheaper/faster models produces commit messages ~50% faster with similar outputs.
commitMessageModelIdsetting toAppSettings(Rust + TypeScript) and threads it through the Tauri/daemon RPC layergetModelListfan-out in SettingsformatModelSlugto produce consistently formatted display names from model slugs (e.g.gpt-5.3-codex→GPT-5.3-Codex)Test plan
vitest— tests foreffectiveCommitMessageModelId,formatModelSlug, andparseModelListResponseshould passDemo
Screen.Recording.2026-02-19.at.2.30.09.PM_compressed.mov