Hyrax AI

Diff-based code review isn't enough anymore. If a tool doesn't run the code and understand the broader system, issues just pile up. The real pain in engineering right now is the lack of continuous architecture adherence. Teams let AI tools generate thousands of lines of fast code that ignore system boundaries and conventions, until a breaking point forces a painful, manual refactor. Hyrax was built to stop that drift before it starts. It keeps a deep architectural map (HYRAX.md) inside the repository, so human developers and AI coding agents work from the same context, and every merge updates it. AI gave software teams velocity. Hyrax helps them ship excellence. Watch the walkthrough below to see how six specialized agents grade code across security, correctness and performance without burying the team in findings. 👇 Get your repository score for free, link in the first comment.

Two exciting changes to Hyrax pricing today, Free just got a lot more generous! 1) Every workspace gets the full product, a $30 signup credit, $10 in fresh credit every month, and up to 100 PR reviews a month. Spend the credit on anything, audits, fixes, or reviews, and no credit card required. 2) Paid is now dead simple. One plan at $30 per user per month, with every feature available. We want as many developers as possible running the full product, seeing it work on their code, and only paying once their team grows. Happy shipping! Try it free: hyrax.dev/pricing

Hyrax free just got a lot more generous, and I'm pumped about it. Every workspace now gets the full product: a $30 signup credit, $10 in fresh credit every month, and up to 100 PR reviews a month. Spend it on audits, fixes, or reviews. No credit card. I wanted anyone to point Hyrax at their repo and see it work before paying a cent. Now you can. Try it free: hyrax.dev/pricing

I compared Hyrax AI against the other tools in this space: CodeRabbit, Greptile, Copilot, Snyk, Semgrep, SonarQube, Qodo. Most are genuinely good at their job, and the breakdown gives credit where it's due. All of them stop at telling you something is wrong. Someone still has to write the fix and get it through CI. Hyrax writes the fix, runs it through CI to confirm it works, and submits the PR. A person still merges it. Full breakdown, tool by tool: hyrax.dev/compare

Introducing the Dispatch. Building Hyrax means keeping up with everything happening in AI and engineering, and the team already publishes what matters on the blog most days. The Dispatch is the weekly version: one email with the incidents, releases, and research that mattered, and what they mean if you ship code with agents. It's free, and the blog reads the same way if you want to see the style before subscribing: https://hyrax.dev/blog

We shipped an MCP server for Hyrax this week. The short version: the coding agents your team already runs, Claude Code, Cursor, Copilot, can now talk to Hyrax directly. Ask which findings touch the file you're about to edit, what conventions apply before you change it, kick off a fix job, check whether the fix PR merged. Same protocol the agents already speak, so there's nothing new to learn. The part I care about most is the live part. We publish each repo's patterns and conventions as files that agents can read, and that works, but files go stale the moment the next audit runs. The MCP server serves the same content live. An agent editing code at 2 PM gets the 2 PM picture of the codebase, not last month's. Same guardrails as everything else we build. Scoped API keys, and a key's scopes are a hard ceiling, even for the account owner. Rate limits per key. Every fix still arrives as a pull request that a human reviews and merges. We don't want LLM stuff running on your code without you knowing what it's doing. Still early. If you hook it up and something's rough, tell me. Docs: https://lnkd.in/gbvyuCgf Write-up: https://lnkd.in/g26tj938