Hyrax is the independent review layer for AI coding. Teams generate more code than ever, much of it written or assisted by AI, and the review step has become the bottleneck. Hyrax takes that step on. It reviews code, finds the issues, writes the fixes, and submits pull requests. The product lives at hyrax.dev.
The work starts with an audit. Hyrax reads the codebase and surfaces real issues rather than style nits: security problems, correctness bugs, performance drags, and the kind of defects that pass a quick human skim. Each finding names the file and line it came from, so a developer can check the claim instead of taking it on faith.
Findings land in a work queue, not an inbox flood. Issues that are cheap to resolve get marked as easy wins, advisories carry guidance rather than a change, and severity ordering puts the findings that matter at the top. A team can go from a fresh audit to a shortlist of real work in one sitting.
Then comes the part that separates Hyrax from tools that only comment. For the issues worth fixing, Hyrax writes the fix and submits a pull request. Tests, build, and lint have to pass before the fix reaches a reviewer. Fixes never merge automatically. A developer reads the pull request, and the change ships only when a human approves it.
The review does not stop after the first audit. Every pull request and commit gets reviewed, so the codebase stays covered as it changes rather than getting one inspection and then drifting.
Getting started costs nothing. Connect a repo, run a mini audit, and start shipping fixes. The free plan includes five findings and five fixes monthly, which is enough to judge the quality of the work on a real codebase before paying anything.
The fastest way to understand Hyrax is to point it at a real repo and read what it finds. The place to do that is hyrax.dev.