open source · MIT

Code review that never sleeps

An open-source PR reviewer that runs on the Claude Code subscription you already have. Every PR gets a full-context review — cloned, read, scored, posted — on your own machine, with nothing extra to pay.

Get started See how it works
$0 / seat self-hosted powered by Claude Code
acme/payments-api #128 REVIEWING · 04:31
10:42:01Cloned repo into throwaway dir
10:42:48Reading diff across 14 files
10:43:37Tracing callers of finish() across repo
10:44:12Flagged unhandled rejection in finish()
10:46:30Drafting review summary…
P0 0 P1 1 P2 2
4.5
› 01 / How it works

Open a PR. The loop does the rest.

Reviewloop watches your repos over webhooks and runs the full review loop — until the PR is verified or merged.

01
PR opens
A webhook queues the review. New pushes dedup by head SHA — no double runs.
02
Claude reviews
Claude Code clones the repo and reads the diff with full codebase context — not just the patch.
03
Findings posted
Issues land on the PR ranked P0–P2, with a 1–5 confidence score on the whole change.
04
Loop until verified
Your agents pick up findings, push fixes, and reviewloop re-reviews — until it's clean.
› 02 / Features

A review console, not a chat window

Your subscription, your keys
Runs on the Claude Code plan you already pay for. No API metering, no extra subscription — and your code never leaves your machine.
Full-repo context
Each review clones the repo and traces callers, types, and tests — it catches what diff-only reviewers can't see.
Agent hand-off
Findings are structured for coding agents. A local Claude Code session can ack a review, fix, and push — hands-free.
Live streaming console
Watch every review as it runs — log lines, elapsed time, status transitions — across all your repos in one place.
Severity that means something
P0 blocks, P1 warns, P2 is a nit. Counts stay dim until non-zero — a quiet console is a healthy one.
Knows when to stop
Merged or closed PRs drop out of the loop automatically. Failed runs retry with a budget, then surface loudly.
› 03 / Why reviewloop

An alternative, not a replacement.

Hosted reviewers like Greptile are polished products, and for plenty of teams they're worth every cent. reviewloop isn't trying to beat them — it's for the rest of us, who'd rather reuse a subscription we already pay for.

A hosted reviewer fits when…
You want a managed service — no worker to run, nothing to update
Reviews should run even when your machine is off
Team dashboards, analytics, and vendor support matter to you
Per-seat pricing fits comfortably in the budget
reviewloop fits when… MIT · self-hosted
You already pay for Claude Code — reviews cost nothing extra
You'd rather your code stay on your machine — a throwaway clone, deleted after
You like owning the pipeline — fork it, tune the prompts, read every line
Your coding agents should pick up findings and push the fixes themselves
› 04 / Install

One prompt. Your agent does the rest.

Paste the install prompt into Claude Code or Codex — it clones the repo, creates your own free Convex backend, starts the worker and console, and watches your first repo. If claude works in your terminal, reviewloop works too.

$ claude # or codex
> Fetch https://raw.githubusercontent.com/fagnersales/reviewloop/master/INSTALL.md and follow it to install reviewloop for me.
✓ watching 1 repo · reviews run on your Claude Code plan