diffray

About diffray

diffray is AI code review built on the argument that the problem with existing AI reviewers is not the AI but the single-agent architecture behind it. The pitch is specific about the failure mode it is attacking: eighteen style nitpicks per pull request that nobody reads, a rename suggestion that misses the SQL injection sitting two lines below, and refactor advice for a pattern the team explicitly rejected three months ago. diffray runs multiple agents over a pull request instead of one model, and it distinguishes between speculating about a change and investigating it — the worked example on the site takes a function-signature change and contrasts the generic single-model response with an investigation that traces the callers. The claimed results are 87% fewer false positives, three times more real bugs and no duplicate comments, all footnoted as internal testing rather than dressed up as independent benchmarks, which is the honest way to publish a number like that. You can try it without an account by pasting any public GitHub pull request URL into the homepage. Integrations cover GitHub, GitLab, Bitbucket and on-premise, and the security posture is stated as SOC 2 infrastructure, GDPR-ready and code never stored. Pricing is per developer with unlimited reviews on every plan, which removes the usual per-review anxiety, and public repositories are free forever. Worth noting: the 26-50 developer band is not on the public page and requires contacting them for a flat Scale price.

Key Features

Tags