What it does
git-lrc adds an AI review step to git. After a one-time setup, it hooks into git commit and examines the staged changes before they enter history, aiming to catch problems that AI coding agents introduce, such as removed logic or changed behaviour. Each review opens in a local browser page that shows the diff with inline comments tagged by severity, a summary, and the list of staged files. Flagged issues can be copied back to a coding agent to fix, and the review repeated. A developer can instead vouch for a commit personally or skip the review. Every commit gets a status line appended to its git log message: reviewed and vouched commits record the number of review rounds and the share of the diff that was reviewed, while a skipped commit is logged only as skipped. Reviews use Google's Gemini API by default. Users can add their own keys for OpenAI, Claude, DeepSeek, OpenRouter or Atlas Cloud, or an Anthropic-compatible endpoint, and set which connector is tried first. The page says only the staged diff is analysed and that diffs are not stored after review. The source is available under a modified Sustainable Use License rather than a standard open-source licence. A separate team product called LiveReview adds dashboards, org-level policies and review analytics.
Capabilities
- Automatic review of staged changes when you run git commit
- Inline comments with severity badges in a browser-based diff view
- One-click copy of flagged issues to paste into an AI coding agent
- Review, vouch or skip outcomes recorded in the git log
- Bring your own key for Gemini, OpenAI, Claude, DeepSeek, OpenRouter or Atlas Cloud
- Hooks can be disabled or re-enabled per repository
When to use it
- Checking AI-generated code changes before they are committed
- Catching leaked credentials or sensitive log output in a diff early
- Giving a team a visible record of which commits were reviewed
- Reviewing an older commit or a range of recent commits
