← Back to Discover

Pi

Open-source, MIT-licensed agent toolkit: a minimal coding agent for the command line that you shape with extensions, skills and prompt templates, plus a unified LLM API, agent runtime and terminal UI libraries.

Coding agentsCommand-line toolsAgent frameworks
Visit website ↗

What it does

Pi is an open-source agent toolkit published by earendil-works under the MIT license. Its main piece is a coding agent CLI that the README calls a minimal, extensible harness: it ships with strong defaults but leaves out features such as sub-agents and plan mode, and you add what you want through extensions, skills, prompt templates and themes, which can be bundled as packages and shared via npm or git. You can use Pi interactively, automate it in print or JSON mode, control it over RPC, or build your own apps with its TypeScript SDK. The monorepo also publishes the libraries underneath: a unified multi-provider LLM API (OpenAI, Anthropic, Google and others), an agent runtime with tool calling and state, a terminal UI library, and a durable conversation and task runtime. For a built-in provider, running /login inside Pi connects a subscription or API key. Pi has no built-in permission system, so it runs with the permissions of whoever launched it. The documentation points to three ways to contain it: a Gondolin extension that routes tools into a local micro-VM, plain Docker, or OpenShell. Install it with the installer script on macOS, Linux or Windows, with npm, or with Nix. Node.js 22.19 or newer is required.

Capabilities

  • Interactive coding agent CLI customized with extensions, skills, prompt templates and themes
  • Use it interactively, in print or JSON mode, over RPC, or through a TypeScript SDK
  • Unified LLM API covering several providers, including OpenAI, Anthropic and Google
  • Separate packages for the agent runtime, terminal UI and durable task runtime
  • Documented sandboxing options: Gondolin micro-VM, Docker and OpenShell
  • Installer that pins dependencies, with npm and Nix as alternatives

When to use it

  • You want a coding agent you can reshape with your own extensions instead of a fixed feature set
  • You are building your own agent app and want an LLM API layer, agent loop and terminal UI as libraries
  • You want to run a coding agent from scripts or other programs rather than only by hand