pi

Why we included this project

Pi is a coding agent that runs in the terminal and can extend itself, with the pieces underneath published as separate packages rather than locked inside one binary. The agent loop, tool calling, and state management are all reusable, so a team can build its own assistant on top instead of being stuck with the CLI. The same repo also ships a unified LLM client that talks to OpenAI, Anthropic, Google, and other providers, which makes switching backends a config change rather than a rewrite. Pi has no built-in permission system, so it runs with the privileges of whoever launched it; the docs cover containerization and sandboxing for teams that want tighter boundaries. That combination of a ready-to-use coding agent and a toolkit you can extend is what makes it worth a look.