Read the change
Diff, related files and the tests that cover them.
Agentic
Code review
with gates
An agent reviews the change and proposes findings with evidence; a reviewer accepts before anything reaches the pull request.
Review agents are widely available; the difference here is governance. Findings are proposals with evidence and a recorded cost, a human reviewer accepts or dismisses each one, and only accepted findings are posted. Dismissals are kept so the reviewer's judgment is measurable.
The pipeline
Propose findings
An agent proposes findings with the lines and evidence behind each, inside a budget per review.
Reviewer accepts
A person accepts, edits or dismisses each finding. Nothing is posted yet.
Post and record
Accepted findings are posted; dismissals are recorded for calibration.
What the record keeps
The change, the proposed findings with evidence, the reviewer's decisions, and the cost of the review.
A fast first pass
Findings with lines and evidence, within a budget per review.
The reviewer stays the author
Each finding is accepted, edited or dismissed by a person before it is posted.
Calibration over time
Dismissals are kept, so the review model's precision is measurable.
Connectors and tool coverage depend on the installed operators and your deployment.
See how failure can trigger a reviewed improvement cycle →Run it on your own infrastructure.
Apache 2.0 licensed, self-hosted. Talk about your review process and what a missed finding costs.