Skip to content

Synced from satyrn-engine 9aecfb513be3211d5f5753bf778807ccecfd76ca; do not edit here.

See also The numbers and the satyrn-engine repository. The engine's terms are in the glossary.

Satyrn Engine

Help Python developers write code their way, using Local AI.

Satyrn is a two-repo effort to make that practical: a developer's own AI partner works on their machine, in their repo, at their pace — and its output arrives as something they review and own, never as a rewrite of their working tree underneath them.

What it owns — and doesn't

The engine owns:

  • contract parsing and validation;
  • contract-aware writable-path and revision enforcement for one replacement;
  • candidate worktree, commit-or-discard, and receipt behavior;
  • the Pi-side guards: loop breaker, writable-path scope, symbol preservation, command bounds — the last three only inside /implement;
  • the Pi package and its thin TypeScript adapter;
  • one real Pi attempt that connects isolation to bounded replacement;
  • the internal Pi-adapter protocol and its compatibility fixtures.

It does not own workloads, grading, repeated runs, comparison statistics, or contract authoring. Those live in the satyrn-evals repository, or stay a main-agent skill. That split is deliberate: evals runs the measurements, and the features built into the engine are the ones that evidence surfaces — no machinery ahead of its contract.

More: glossary — the terms used here (contract, adapter, protocol, refusal, …), defined in this repository's own words.

Release one ships /implement: a developer-invoked bounded operation that runs one piece of work in a fresh model context and its own worktree under a derived contract, with the guards loaded, and returns a candidate plus a compact receipt. Design and roadmap: the satyrn-evals repository, docs/superpowers/specs/2026-09-13-release-one-design.md.

License

Apache-2.0 — see LICENSE.