Review modes
Last updated: August 2026
Every review runs in one of three modes. Deeper modes read more context and cost more review units — your plan includes a monthly unit allowance (e.g. Pro includes 150).
Quick — 0.5 units
Checks: Only the changed diff, for the things that break production: obvious functional bugs, security issues, severe performance problems, data-loss risk, and breaking behavior.
Use it for: High PR volume, small changes, pre-merge sanity checks.
Standard — 1 unit
Checks: Everything in Quick, plus maintainability, error handling, framework-specific issues (React, React Native), TypeScript issues, API misuse, missing edge cases, and likely missing tests.
Use it for: The default. Right for most day-to-day PRs.
Deep — 2 units
Checks: Everything in Standard, plus related unchanged files as context — interfaces and types, call-sites, project config and rules — to catch architecture problems, backward-compatibility breaks, and contract implications.
Use it for: Risky changes: refactors, API/database contract changes, release branches.
How the default mode is chosen
Per repository, in order of precedence: a one-time choice when you submit a PR URL → dashboard repository settings → .ai-review.yml → your workspace default (Standard). Regardless of mode, findings pass the same validation pipeline — confidence gates, deduplication, and diff-line mapping — before anything is posted.