Autonomous code repair starts with understanding.

Autonomous code repair starts with understanding.

An agent that reads your repo through the model spends most of its tokens finding the bug. Codna maps the repository first, without a model, and hands the agent an evidence bundle scoped to the issue. It prints the raw-to-bundle size on every run.

The result is a fix that explains itself: root cause, confidence, blast radius and regression risk, gated by a risk check before anything is applied. Run it with --tests --apply and Codna re-runs your tests until they pass.

Core pattern

1. Map the repo for zero model tokens.
2. Identify suspect files and call paths.
3. Pack an evidence bundle scoped to the issue.
4. Generate a patch and report root cause, confidence and risk.
5. Gate it, test it, and open a pull request you review.