$ gitclaw review --pr 482

An AI reviewer that reads your diffs — not your code.

GitClaw is an open-source, self-hosted alternative to CodeRabbit. It runs on your own infrastructure, reviews pull requests on GitHub, GitLab, and Bitbucket, and leaves inline comments on the lines that actually matter — security, performance, maintainability. Reviews use your chosen AI provider, including a local model.

MIT licensed · see all releases
pull_request #482 · payments-service
14+ await db.query(`SELECT * FROM users WHERE id = ${userId}`)
⟫securityUnescaped userId goes straight into the query string — this is injectable. Use a parameterized query instead.
31+ items.forEach(i => results.push(transform(i)))
⟫performanceThis rebuilds results on every render. items.map(transform) avoids the mutation and reads cleaner too.
58+ @gitclaw why flag this as high severity?
⟫This table holds payment metadata — an injection here reaches billing data, not just user profiles. That's why it's high, not medium.

WHY TEAMS RUN THEIR OWN

Eight things it actually does

Not a feature wishlist — this is what ships in the current release.

⟫ 01

Reviews on every push

Webhooks fire on open and update. GitClaw diffs only what changed, so a follow-up commit doesn't re-review the whole PR.

⟫ 02

Security & code quality

Flags injection risks, unsafe patterns, and style violations before they reach main — not after.

⟫ 03

Performance insights

Catches hot-path inefficiency, redundant work, and regressions that quietly slow a service down over time.

⟫ 04

Maintainability feedback

Calls out complexity and duplication while it's still cheap to fix — not eighteen months later.

⟫ 05

Talk back to the review

Reply to @gitclaw directly in a PR comment. It answers follow-up questions in context, no new tab required.

⟫ 06

Per-repo configuration

Drop a .gitclaw.yaml in any repo to set ignore paths, review tone, and custom instructions.

⟫ 07

Dashboard & analytics

Track findings across every repo and team in one place. See what gets caught, and what keeps recurring.

⟫ 08

Bring your own model

Pluggable AI backend — OpenRouter, Anthropic (Claude), Groq, or any OpenAI-compatible endpoint, including a local Ollama instance.


YOUR SERVERS, YOUR CODE

Self-hosted means self-hosted

GitClaw doesn't send your source to a third-party API to get reviewed. It runs wherever you already run things, talks to the model provider you choose, and keeps every diff inside infrastructure you control. The dashboard is yours; the data never leaves.


DESKTOP APP

Download for your machine

Native installers with embedded Postgres — no Docker required. Hosted on gitclaw.online when available.

Desktop builds

Download GitClaw for Windows, macOS, or Linux. Choose the installer for your machine.

These installers are unsigned. Windows and macOS may display security warnings; the macOS app is not Apple notarized. Download only from this site or the linked GitHub release.

View release notes and all available installers on GitHub

Before you install

Node.js, Docker, Git, and a separate database are not required for local reviews. GitClaw includes its own local PostgreSQL database.

On a Mac, open Apple menu → About This Mac. Choose Apple Silicon if you see an Apple M-series chip, or Intel if you see an Intel processor.

Linux installation commands

Run these commands in the folder containing your download.

Debian / Ubuntu (recommended)

sudo apt install './GitClaw-0.1.2-amd64.deb'

AppImage

chmod +x './GitClaw-0.1.2-x86_64.AppImage'
'./GitClaw-0.1.2-x86_64.AppImage'

If AppImage reports a missing FUSE library, install FUSE 2 using your distribution’s package manager, or use the .deb on Debian / Ubuntu. Launch the app as your normal user, without sudo.

Your first review

  1. Open GitClaw. The first launch initializes the local database and may take about a minute.
  2. Choose an AI provider and enter its API key, or connect a running local model server and specify its model name.
  3. Open Local review, choose a source folder, and inspect the included files and AI destination.
  4. Confirm the review to send the selected code to your chosen model and see findings.

Cloud models need internet access and may charge for API usage. Your source files are not executed or modified during a local review.

GitHub, GitLab, and Bitbucket connections are optional for local reviews. For automatic pull request reviews, configure a provider and public webhook URL in Integrations, and keep GitClaw running.


GET STARTED

Point it at a repo. See what it finds.

Five-minute setup, MIT licensed, no usage limits because there's no meter to hit.