Hyrax

Hyrax is a coding tool. Hyrax connects to GitHub, maps a repository into in-repo context files, and uses six specialized agents to review pull requests and open verified upgrade PRs. Engineers approve and merge every change; Hyrax never merges on its own.

Hyrax is a codebase architecture and pull request review tool built for teams that write a lot of code with AI assistants. It connects through a GitHub App, reads an entire repository, and writes a map of it back into the repo as HYRAX.md and a .hyrax/discovery folder. That folder holds architecture notes, environment variables, patterns, principles and how-to guides, so human developers and AI coding agents work from the same shared context. Once the map exists, six specialized agents evaluate the repository across security, correctness, maintainability, performance, architecture and operations. Findings appear with stable HYRAX-N references, severity, file location and an effort estimate, and the team can approve a fix, dismiss it or request changes. Approved work is carried out in an isolated Git worktree and passes a 13-step verification gate that includes baseline tests, regression checks, build, lint, formatting, a diff size limit, a second review by an independent agent and a re-scan confirming the issue is gone. If a required check fails, no pull request is opened. Hyrax never merges on its own, so an engineer reviews every change. Pull requests on connected repositories also receive a summary comment and a GitHub check, and a newer change review covers a whole change across several pull requests and repositories. The product suits engineering teams that already use assistants such as Cursor, Copilot, Claude Code, Codex or Devin and want a layer that tracks the overall system rather than a single edit. It also offers Linear and webhook triggers, benchmark scores that track repository health over time, and a stated policy of running inference on AWS Bedrock without training on customer code. Compared with pure code generation tools, Hyrax sits closer to automated review and technical debt management. It overlaps with PR review bots, but its persistent in-repo context, upgrade pull requests and verification pipeline set it apart. Teams wanting only inline comment suggestions may find it broader than necessary.