Skip to content

About Felix

Felix is a delivery harness for AI coding agents. It exists because capable coding agents still need structure: a written plan to work against, tests that gate progress, and durable records of what happened.

How Felix works

  1. Spec to plan. Felix writes an explicit implementation plan from your spec before any code changes. You can inspect it, reject it, or let it proceed.
  2. Plan to code. The agent of your choice implements one task per iteration against the plan.
  3. Code to validation. Tests, builds and lint checks act as backpressure. Nothing is marked complete until they pass.
  4. Validation to commit. Only validated work is committed, and all state lives in files you can version and audit.

Principles

  • Durable state on disk, not chat history.
  • One task per iteration: an atomic unit of progress.
  • Backpressure from tests and builds is non-negotiable.
  • Specs are the persistent test suite; plans are ephemeral to-do lists.
  • Your code stays on your machines; sync is optional.

Source, releases and contact

Felix is developed in the open at github.com/nsasto/felix. Questions and bug reports are welcome as GitHub issues. See the changelog for release history, the documentation to get started, and our privacy policy and terms.