AI Review

Key features

What Angada's AI review does on every pull request.

Automatic review on every pull request

Angada reviews each pull request the moment it's opened. Push more commits and the review continues — Angada waits out a 90-second debounce so a burst of pushes becomes one pass, then reviews what changed. You never configure webhooks or CI steps for this; installing the GitHub App is the whole setup.

A code finding connected to its review evidence.
A reviewed pull request

Open PR with an Angada review check and one inline finding comment visible.

A verdict on every PR

Each review ends in one of five states, posted straight to the PR's checks:

VerdictMeaning
ReviewingThe review is still running
ClearNothing at or above your severity floor was found
Needs humanFindings were posted for a person to weigh
BlockedThe review couldn't run — for example, the PR exceeded the 150-file ceiling
SkippedThe PR didn't qualify (draft, bot author, or a cap was hit)

The verdict tells your team whether a PR is clear or needs a human pass.

A review signal beside the changed code that produced it.
Verdict states on the checks list

The Angada check in Reviewing, Clear, and Needs human states.

Findings with severity, confidence, and evidence

Every finding carries a severity (high, medium, or low), a confidence score, and the exact file and lines it concerns — quoted so you can judge it without leaving the comment. Findings at or above your Workspace's severity floor (default: high) land as inline comments; the tier below is summarized; lower still, they're stored in Angada without touching the PR. See Understand a review for the full anatomy.

A code finding connected to its review evidence.
An inline finding with severity and confidence

One comment showing severity badge, confidence, evidence quote, and fix instruction.

Fix-with-agent handoff

When a finding needs code changes, each review comment carries a handoff badge for Cursor, Claude Code, or Codex. Click it and Angada copies a ready-to-paste prompt containing the finding and its evidence. The handoff does not commit to the pull request. See Fix with a coding agent.

A review finding flowing to a coding-agent handoff.
Fix-with-agent badge to Cursor

Short clip: click badge on a finding, paste prompt into the agent, agent opens a fix.

Draft-aware and bot-aware

Draft PRs are left alone until they're marked ready for review — no wasted reviews on work-in-progress. Pull requests from bots are skipped entirely. Both behaviors are defaults, and draft handling is configurable in Review settings.

Re-runs that respect your access rules

If a PR's head branch loses write access mid-flight, Angada re-runs the review rather than posting a stale verdict. A user with write access can request a manual re-run. It bypasses the 150-file automatic-review ceiling but remains subject to the rolling review caps.

GitHubPull request activity
AngadaReview and coordination
SlackLiving pull-request card
Requesting a manual re-run

The re-run action on an Angada review check.

Built-in limits that keep reviews predictable

Three reviews per PR and ten per Workspace in any rolling 24 hours, by default. Hitting a cap marks the PR skipped rather than queueing silently. Caps and the severity floor live in Review settings; the full rules are in Limits and caps.

On this page