General-purpose AI reviewers get noisy fast. PR Quorum splits the job into Correctness, Security, and Architecture, then lets each reviewer keep its own model, focus list, confidence threshold, and posting behavior.
Specialists
3
Model routing
per reviewer
Policy
per repo
Default cap
10 comments
Default reviewers
Correctness
Finds likely bugs, regressions, edge cases, and test gaps by arguing backward from runtime failure modes.
Product code, auth code, migrations, and generated files should not get the same review. Save a custom panel once, then override reviewer behavior where the repo demands it.
Each reviewer returns structured findings with severity, confidence, file, line, title, body, and an optional suggestion. That structure is what lets PR Quorum filter noise before it reaches the PR.
1. System prompt
Reviewer focus list, merged with your .ai-review.yml policy
2. PR diff
Sent verbatim to OpenRouter chat-completions
3. Structured output
Zod-validated JSON, parsed in the Inngest function
4. Confidence floor
Drop anything below min_confidence (default 0.75)
Finding · structured output
{
"severity": "high", // low | medium | high | critical
"confidence": 0.87, // 0..1, dropped below min_confidence
"file": "src/billing/stripe-webhooks.ts",
"line": 142,
"title": "Webhook signature verified after side-effects",
"body": "...prose for humans...",
"suggestion": "...diff-shaped fix..." // optional, posted as GH suggestion
// reviewerId is added by the runtime
}
Put the specialist panel on your next PR
Start with the default three reviewers, then tune the model, confidence, and focus as your repo grows.