Hexmos' git-lrc bet: commit-time AI review catches what PR review misses

git-lrc was born from a felt problem at Hexmos, the maker of developer tools behind the LiveServer extension family: in 2025, as the team leaned harder on AI coding tools, it generated far more code but spent less time reading it. Regressions occasionally slipped through, and some changes made it all the way to production only to be reverted later. Existing AI code review tools all operated at PR time, which the author, engineer Shrijith Venkatramana, considered too late — by then the faulty code was already committed, pushed and visible.

The product attaches to git commit itself: a hook opens a review UI with the diff before the commit lands, the AI flags risky changes inline, and the developer can fix, vouch for, or consciously skip — each decision recorded in git history. git-lrc was pitched as the braking system for AI code, since AI agents can silently remove logic, relax constraints, add expensive cloud calls or leak credentials. It launched on Product Hunt on 2026-02-21 as free and unlimited, running on bring-your-own model keys with Gemini default, and reached #3 Product of the Day with 318 upvotes and 28 comments.

Over the following months the tool expanded from a simple review checkpoint into a fuller product: about 100 failure patterns across 10 risk categories, a risk-scored view that ranks hunks by blast radius, version-controlled .lrc/ repository rules that teach the reviewer a team's conventions, and claude-lrc, which brings the same review-vouch-skip workflow into Claude Code. Pricing also evolved from the launch promise of unlimited free reviews to a LOC-based model — 30,000 reviewed lines per month free, premium from $32 per 100,000 lines — as the company found a sustainable monetisation shape.

As of September 2026 git-lrc remains live and open-source under HexmosTech on GitHub, with a free individual tier, premium and enterprise tiers, and no disclosed revenue or funding. The bet is that commit-time review becomes a normal engineering habit for AI-assisted development, rather than another dashboard teams ignore — still unproven at scale, but the launch numbers showed the problem resonates.