Find bugs before your coding agent opens a pull request
Perch on Greptile's 50 pull requests and ReviewBench: recorded bug detection, precision, thresholds, model costs, and scripts to reproduce the scores.
Perch / Blog
Blog
Building Perch, catching bugs, and checking the code our agents write.
Stop AI slop PRs before they reach code review
Coding agents cheat on tests, report fixes that didn't happen, and swallow errors. Here's a real swallowed error, and the plain-English Perch rule that fails the PR.
Can Jev find bugs and security vulnerabilities?
Benchmarking Jev and other decision models on code fixes and semantic lint rules
Fixing AI Landing Pages with Jev
I used Perch to check the landing-page habits I kept asking Codex to fix.
Semantic linting with Jev
Example: how Perch works on a semantic lint rule
Introducing Perch
A semantic linter for bugs, security, and project-specific rules