It speaks your dialect.
Some teams say chore for CI, others say ci. git guess indexes your last thousand commits and looks at how you named the closest changes. It suggests your convention, not a generic one.
It reads the diff and writes the Conventional Commits type, the way your repository would. One binary, no API key, nothing leaves your machine.
brew install Halleck45/tap/git-guess
curl -fsSL https://raw.githubusercontent.com/Halleck45/git-guess/main/install.sh | sh
npx git-guess
go install github.com/Halleck45/git-guess/cmd/git-guess@latest
Linux, macOS, Windows. Binaries if you prefer. Or try it in your browser first.
$ git guess # type for the staged changes
$ git guess -m "add login" # feat(auth): add login
$ git guess eval # score it on your own history
$ git guess hook install # git commit -m "add login" becomes feat(auth): add login
Paste git show or git diff. The model runs in your browser. Or pick one:
This is the model alone. Installed, it also reads your repository's history, which is where most of the accuracy comes from.
Some teams say chore for CI, others say ci. git guess indexes your last thousand commits and looks at how you named the closest changes. It suggests your convention, not a generic one.
Above 90% confidence it is right 96% of the time. Below 55% it asks, once, in one keystroke. The hook never blocks a commit.
? fix 41% or refactor 38%? [f/r]
One line in a workflow. Every PR gets enhancement, bug or documentation, and a note when a commit's declared type contradicts its code.
- uses: Halleck45/git-guess@v1
70% right on the first try, 88% within two, on 36 projects it had never seen. Trained on 488,000 commits from 244 open-source projects. Your own number: git guess eval.