Pi is an agent harness and self-extensible terminal coding agent. The monorepo contains the interactive coding-agent CLI, an agent runtime with tool calling and state management, a unified multi-provider LLM API covering OpenAI, Anthropic, Google, and other providers, a terminal UI library with differential rendering, and vendor-neutral telemetry contracts. It is also the engine underneath OpenClaw.
Pi has no built-in permission system: by default it runs with the permissions of the user and process that launched it. For stronger boundaries, three containerisation patterns are documented: a Gondolin extension that keeps pi and provider auth on the host while routing built-in tools and ! commands into a local Linux micro-VM, plain Docker for the whole process, and OpenShell for a policy-controlled sandbox. A separate pi-chat project covers Slack and chat automation.
| Package | Description |
|---|---|
| @earendil-works/pi-coding-agent | Interactive coding agent CLI |
| @earendil-works/pi-agent-core | Agent runtime with tool calling and state management |
| @earendil-works/pi-ai | Unified multi-provider LLM API |
| @earendil-works/pi-tui | Terminal UI library with differential rendering |
| @earendil-works/pi-telemetry | Telemetry contracts, reference adapter, conformance tests, and typed schemas |
Features
- Coding agent CLI: interactive terminal agent that can extend itself, including custom workflow extensions with their own TUI
- Agent runtime: tool calling and state management as a reusable package
- Multi-provider LLM API: one interface to OpenAI, Anthropic, Google, and other providers, with generated model data refreshed at build time
- Terminal UI library: differential rendering for TUI components
- Telemetry contracts: vendor-neutral schemas with a reference adapter and conformance tests
- Sandboxing patterns: Gondolin micro-VM extension, plain Docker, or OpenShell
- Supply-chain controls: exact-pinned dependencies, lockfile as ground truth, shrinkwrapped CLI package, --ignore-scripts installs, and scheduled audits
- Standalone binaries: reproducible builds from the versioned release source archive with offline model data
Extensions & skills
Agenta
Workspace for building agents through chat and running them in the background, with swappable harnesses and models
BrowserSkill
Connects shell-capable AI agents to your logged-in browser, running tasks in a separate window without lock-in
Computer Use Linux
Linux desktop control over MCP — AT-SPI, GNOME Shell, Wayland portals, ydotool
Cotal
The open standard for agent coordination
Built with this
Integrates with
Integrated by
Acpx
Headless CLI client for stateful Agent Client Protocol (ACP) sessions; drives coding agents like Codex programmatically.
Agent of Empires
Manage multiple Claude Code, OpenCode agents from either TUI or Web for easy access on mobile. Also supports Mistral Vibe, Codex CLI, Gemini CLI, Pi.dev, Copilot CLI, Factory Droid Coding.
AI DevKit
The control plane for AI coding agents.
Caveman
Skill for Claude Code and 30+ agents that compresses replies into terse caveman-speak to cut output tokens
