Fred

Fred started in July 2026 as an answer to a problem we kept running into on our own repositories at Raw Power Labs. Reviews were where work piled up, and a large share of what reviewers spent their time on was mechanical: things that were obvious, repetitive, and already written down somewhere in our standards.

Fred is a review engine that reads a diff before a human does. It runs several different AI families over the same changes, because their perspectives differ enough that one will catch what another misses, and it grounds every review in the project's own architecture and code standard documentation so its findings match the conventions the team actually agreed on.

Reviews can go further than reporting. A fix loop will attempt the repair and review the result again across a limited number of rounds, and each run is settled as clean, disputed, or failed. Every review carries its duration and its cost in dollars, which was a deliberate decision: an agent that can loop by itself needs a visible price.

As a full-stack project it covers the interface and everything behind it, including the settings that make it configurable per team: which agents run, which providers back them, which repositories are watched, and the learnings that carry forward from one review to the next.

The goal was never to replace review. It was to make it faster and more effective, and to keep a codebase continuously at the standard you set rather than letting it drift between audits.