Pre-push · local · autonomous

Your PR queue should not be your first quality gate.

RefactorAI enforces compliance at commit/push time, reducing review churn and rework.

$ pip install refactorai-cli
Reviews on commit, not after push <100ms hook — never blocks the commit Nothing leaves your machine
The loop

Commit → review → resolve → push

RefactorAI runs a tight inner loop on your machine. The PR is clean before it exists.

1

Commit as usual

You or your agent runs git commit. A non-blocking post-commit hook enqueues it in under 100ms — the commit never waits on review.

2

Reviewed locally

RefactorAI scopes to the changed files plus the code they touch, guided by your commit intent, and raises Refactor Requests.

3

Resolve in the terminal

In the always-on Refactor Terminal, press [r] to apply behavior-preserving fixes and [d] to diff — before push.

4

Push through the gate

The pre-push gate runs warn or strict. Only commits with a passing gate go out — with an escape hatch when you need it.

Why RefactorAI is different

Not another PR-comment bot

Every other AI reviewer fires after you push — a webhook leaves comments on a PR that already exists. RefactorAI moves review left, all the way to git commit, and fixes instead of comments.

Post-push PR review tools
RefactorAI
When it runs
After push, at PR time
On every local git commit
Where it runs
Cloud webhook
Local — no network on the commit path
What you get
Comments on your PR
Behavior-preserving fixes applied
Effect on PRs
Adds a review round
Removes it — PRs arrive clean
Gate
Advisory only
Pre-push gate (warn / strict)

AI writes more code than any team can review. We move review left — all the way to the commit — so your PR queue stops being the bottleneck.

Built for the agent era

Everything happens in one local loop

A commit trigger, a live terminal, and a pre-push gate — designed for developers whose agents write faster than anyone can review.

Commit-triggered review

A post-commit hook reacts to every commit and returns in <100ms. It only enqueues — it never blocks or fails your commit.

The Refactor Terminal

One always-on window beside your editor. Every commit is a node in a live, Graphite-style graph — and it auto-opens after your first commit.

Pre-push gate

Enforce warn or strict before code leaves the machine — with an escape hatch and a fail-open timeout so a slow run never wedges a push.

Fixes, not comments

Actionable findings become Refactor Requests you resolve in place. Every refactor is behavior-preserving; behavioral gaps escalate instead of shipping.

Intent-guided scope

Review follows the change: the files you edited, expanded along the symbol graph to their callers, callees, and impacted modules — not the whole repo.

Agent-native

A --json event stream lets Cursor (or any agent) commit, read review status, self-correct, and push — fully headless.

Who it's for

For teams shipping agent-written code

When agents write most of the diff, the reviewer is the bottleneck. Move the first review pass onto the machine that wrote it.

Developers using coding agents

Cursor writes it and commits it — RefactorAI catches it before it's your teammate's problem, and before it's a PR.

Platform & team leads

Set the team-wide pre-push policy (warn or strict) so quality is enforced at the source, consistently.

Regulated environments

Review runs locally with no network round-trip on the commit path — no cloud copy of your source to leak, subpoena, or breach.

Privacy by architecture, not by promise.

Most AI review tools ship your source to a server and comment on your PR. RefactorAI flips that: the review runs on your machine, on commit, with no network round-trip on the commit path.

Try it now
  • Source code and index never leave your device
  • No network round-trip on the commit path
  • Bring your own key or run a local model
  • Fail-open by design — review never wedges your git workflow

Stop reviewing AI code in the PR. Catch it at the commit.

Install the CLI, open the Refactor Terminal, and push only what passes the gate.