Skip to content

About letscode

  • Changelog

    Per-version breakdown of what shipped.

  • Roadmap

    Where the project is going, and the headlines per milestone.

  • Lessons learned

    What we got right, what we got wrong, what we'd do differently.

  • Related projects

    How letscode relates to Claude Code, pi.dev, and the agent landscape.

Background

letscode is a small Python coding-agent harness: an event-driven async agent loop, a basic terminal frontend, and a pluggy-based plugin system. The core stays minimal — anything that isn't load-bearing for the agent loop or the basic UI is delivered as an installable Python plugin.

For how letscode relates to other agents in the space (Claude Code, pi.dev, OpenAI's tools), see Related projects.

The project follows three working agreements:

  1. Test-drive between milestones. Every release ends with a manual Fireworks/MiniMax run that surfaces UX bugs the automated suite can't catch.
  2. Defer with rationale. Items punted to the next release get a "why" and an unblock condition — not a silent drop.
  3. Write design docs before implementation; validate them with a real consumer. v0.3's extension model is the canonical example: a design doc was written, a reference plugin was written against it to validate the contract, then the authoring guide was refreshed.

License

MIT.

Contributing

The project is currently solo-driven. Issues, PRs, and design feedback welcome via GitHub.