HeyEddi CI

Engineering managers and team leads cannot personally review every pull request, yet merge risk accumulates when review depends on someone babysitting each diff. Existing AI review bots add noisy comment streams that developers tend to reject, and seat-based pricing does not match how teams actually use review tools. Leads also lack visibility into where their review process itself is weak, such as backlog, coverage gaps, or excessive re-reviews, without resorting to ranking individual people.

HeyEddi CI posts a first-pass, evidence-bound review on GitHub pull requests, with findings required to cite a specific file and line or a CI check before they can ship. It runs cited lenses (bug, perf, design) plus an optional security pass over the diff, using a bounded related-context pack built from the import graph rather than crawling the full repo, and redacts obvious secrets and PII-like patterns before calling the model. Teams configure review budgets, external-PR rules, and safety defaults through an eddi-ci.yaml policy file, and interact via slash commands like /heyeddi review, /heyeddi again, /heyeddi fails, and /heyeddi ask. All suggestions remain advisory GitHub suggestions; a human must apply and merge, and the agent never merges on its own. A Free Health feature reports system-level review metrics such as backlog and re-review rates without ranking individuals.