AI CODE REVIEW Open source · Self-hostable
Push your code. Get a review that's actually right.
Cerberus reviews every pull request the moment you open it — full-repo context, not just the diff, and every finding double-checked before it reaches you. Less noise, fewer things to second-guess.
Free to install, no credit card. The free plan includes 30 AI runs a month on a shared model key.
View on GitHub- 01Read full-repo context
- 02Draft findings
- 03Verify every finding
- 04Post to the PR
-
BUG
Potential null pointer at
Verifying… Verifieduser.settings— add null check before access. -
PERF
N+1 query in loop at line 45 — consider batch fetching.
Verifying… Verified -
STYLE
Verifying… DroppedUnused importosin utils.py — remove it.It's re-exported through
__all__— didn't hold up, so it was never posted. -
QUALITY
Good use of early returns. Clean separation of concerns.
Verifying… Verified
Inside the dashboard
Every review, fully traced.
What changed, what was flagged, where each finding sits in the code, and what the run cost — the same review that lands on your pull request.
Signal, not noise
Most AI reviewers optimize for finding things. Cerberus optimizes for finding things that are actually true.
Three safeguards
Not just another noisy AI reviewer.
See the full precision pipeline-
Self-verified findings
Every issue Cerberus raises gets a second, independent look before it's posted — a dedicated verification pass reads the same code and asks 'is this actually true?'. Findings that don't hold up get quietly dropped, not shown to you.
-
Learns your team's conventions
When your team replies "that's intentional" or "that's our convention" on a review comment, Cerberus remembers — and stops raising it on every future PR. The more you use it, the quieter it gets.
-
Steerable, right from the PR thread
Reply
/cerberus ignore,focus, orpauseto redirect a review mid-PR — plus path-specific rules in.cerberusconfigand existing lint results all feed into what it flags.
One platform, three heads
- Review · Stable
- AI code review on every pull request — this page is mostly about it.
- Code · BETA
- An autonomous coding agent that picks up tasks and opens pull requests.
- Hacker · EXPERIMENTAL
- Dependency vulnerability monitoring that opens a GitHub issue for each hit.
Getting set up
Up and running in three steps
-
Sign in with GitHub
We read your GitHub username and email. This does not give Cerberus access to your repositories yet.
-
Install the GitHub App
On GitHub's install screen, choose your account or organization and which repositories to allow. GitHub lists the permissions before you confirm.
-
Switch on your repos
Back in Cerberus, turn on the repositories you want reviewed. Then open a pull request and Cerberus comments on it.
Prefer to trigger it yourself? Comment /review on any open PR to kick off (or re-run) a review on demand.
Installing on an organization? GitHub only lets organization owners install apps; members send an install request to an owner. Read the setup guide
Where does your code go?
- Your code
- Cerberus fetches the diff and the files it needs from GitHub for each review and sends them to the AI model provider you've configured. It keeps review results and run logs, not a mirror of your repository.
- AI usage
- Cerberus supplies the AI key. Runs count against your plan's monthly AI-run allowance; you never hand over a key of your own.
- Your infrastructure
- Open source and self-hostable — run it on your own infrastructure if you prefer.
Your next PR deserves a better review.
Free to install. Set up takes a few minutes. The first review will surprise you.
Get Started with GitHub