MythosReview
AI review on GitHub pull requests. A second model from a different family confirms each finding before it is posted.
A GitHub App reviews a pull request when it opens and when it changes: one summary plus inline comments with a severity from P0 to P3, and a chat reply when you mention it. Context is the repository, not only the diff.
The difference is a verification pass. A second, different-family model confirms that each finding is real, and where a check command is configured it runs once and a finding whose prediction does not match is dropped. Unconfirmed findings are not posted.
A per-repository Learnings file is editable by the team, so a dismissed finding stays dismissed. Chat never changes code.
- Same family refused
- Using one model family for both roles is a configuration error and the run posts nothing.
- Measured signal
- The report exposes the false-positive rate before and after verification when labels exist. Round one is a fixture, not a production rate.
The pipeline is provable offline against scripted models and a fake GitHub. The live app is not open to the public.
Get early access to MythosReview
This is the first product to open. Tell us where you review code.