PR Quorum — AI Code Review Panel

2026 — ongoing

2026

A panel of specialized AI reviewers (correctness, security, and architecture) runs on every pull request and posts one combined, actionable GitHub review. Install the app, open a PR, get a verdict.

I designed, built, and run PR Quorum end to end: the product, the GitHub App, the review pipeline, billing, and the self-hosted infrastructure it runs on.

Three specialist reviewers working on the same pull request

One review, not 34 comments

Most AI reviewers bury the one finding that matters under a pile of nitpicks. PR Quorum splits the job between specialists, then merges, de-duplicates, and ranks what they found, so a team reads one summary with inline comments on the exact diff lines. Reviews stay advisory: people keep merge control.

How it works

A GitHub App webhook queues a durable Inngest workflow per pull request. Reviewers run in parallel, each with its own prompt, focus, and model through OpenRouter.
Reviewers must quote the line they flag; the quote is checked against the diff, and a verifier pass fact-checks findings before anything is posted.
Deterministic scanners run next to the models: secrets, supply-chain checks on lockfiles and CI workflows, and a prompt-injection tripwire.
Per-repo rules in .ai-review.yml, bring-your-own OpenRouter key, per-reviewer cost tracking, and hard monthly limits keep it tunable and predictable.

Critical and high-severity findings caught before merge

Shipping it

PR Quorum is live with paid plans and a free tier of 50 reviews a month. It runs on a single EU server under Coolify, with a self-hosted Inngest worker built for reviews that legitimately take minutes on large pull requests.

One combined verdict posted on the pull request