Reviewer LibAI code review for pull requests

Structured, actionable findings from OpenAI — use it as a library, a CLI, or a GitHub Action. Works with code in any language.

Get Started

API Reference

View on GitHub

🎯 Structured findings

review() returns typed Finding[] — severity, category, file, line, message, suggestion — ready for inline PR comments or a CI gate. Not free text.

🧠 Reasoning model by default

Ships with o4-mini for the fewest false positives; it reasons about whether a concern is real instead of hedging. Pass gpt-4o-mini for a cheaper, noisier review.

💬 Smart PR comments

Posts inline comments (that go “outdated” on push) plus a single self-updating summary comment — no stale pile-ups.

📝 Change overview

Opt-in “what changed” summary of the diff, rendered above the findings in the same comment.

⚡ CLI & GitHub Action

Review from the terminal or wire up a PR check in a few lines — no glue code. Fetches the PR diff and posts results back.

🧩 Diff-aware

Annotates added lines so findings anchor to real line numbers; large diffs are chunked by file and hunk, then merged.

🚫 Noise control

Exclude path globs (lockfiles & build output skipped by default), a severity gate (--fail-on), and an optional second-pass triage.

♻️ Reliable & typed

Configurable retries and timeouts, a result cache to skip unchanged input, npm provenance, and shipped TypeScript types.