Reviews what matters. Skips what doesn't.

Most review tools send your entire diff to an LLM and hope for the best. inspect understands your code structure first. It parses every PR into functions and classes, maps their dependencies across files, and scores each change by how much damage it can do. The LLM only sees what actually matters.

How it works

1. Your PR gets parsed into entities (functions, classes, methods)
2. A dependency graph maps what calls what, across every file
3. Each entity gets a risk score based on blast radius, change type, and dependent count
4. Only the high-risk entities go to the LLM for deep review
5. You get precise findings on the changes that can actually break things

Entity-level triage

Your PR touches 100 entities. 8 of them modify public APIs with 40+ downstream dependents. The other 92 are formatting, comments, and leaf functions nobody calls. inspect finds the 8 that matter and ignores the rest. 92% less noise. No bugs missed.

Blast radius analysis

A one-line signature change looks harmless in a line diff. inspect sees it breaks 47 callers across 12 files. Every finding comes with context: what depends on this, what breaks if it's wrong, and why it was flagged.

21 languages

Rust, Python, TypeScript, Go, Java, C, C++, C#, Ruby, Swift, Kotlin, Scala, PHP, Haskell, Lua, Zig, Elixir, Dart, Vue, Svelte, ERB.

Built for speed

Entity extraction and graph building run locally in milliseconds. No waiting for LLM roundtrips on code that doesn't matter. Reviews come back faster because the LLM is only reading what's important.