Pi Coding Agent

Pi is a minimal terminal coding agent designed to adapt to your workflows rather than forcing you to adapt to it. Built as an extensible harness, Pi provides powerful defaults while allowing developers to customize and extend functionality through TypeScript extensions, skills, prompt templates, and themes. The tool supports 15+ AI providers including Anthropic, OpenAI, Google, Azure, Bedrock, Mistral, Groq, and more, with authentication via API keys or OAuth.

• Extensible Architecture - Build custom features using TypeScript extensions with access to tools, commands, keyboard shortcuts, events, and the full TUI. Create sub-agents, plan modes, permission gates, path protection, SSH execution, sandboxing, and MCP integration.

• Multiple AI Providers - Connect to Anthropic, OpenAI, Google, Azure, Bedrock, Mistral, Groq, Cerebras, xAI, Hugging Face, Kimi For Coding, MiniMax, OpenRouter, Ollama, and more. Switch models mid-session with /model or Ctrl+L.

• Tree-Structured Sessions - Sessions are stored as navigable trees. Use /tree to jump to any previous point and branch from there. Export to HTML with /export or share via GitHub gist with /share.

• Context Engineering - Control what goes into the context window with AGENTS.md for project instructions, SYSTEM.md for custom system prompts, auto-compaction for older messages, and skills for on-demand capability loading.

• Skills System - Capability packages with instructions and tools loaded on-demand for progressive disclosure without busting the prompt cache.

• Prompt Templates - Reusable prompts stored as Markdown files that expand with /name command.

• Message Queuing - Submit steering messages while the agent works (delivered after current tool) or follow-up messages that wait until the agent finishes.

• Four Integration Modes - Interactive TUI, print/JSON for scripts, RPC via JSON protocol over stdin/stdout, and SDK for embedding in applications.

• Package System - Bundle and share extensions, skills, prompts, and themes as packages. Install from npm or git with version pinning support.

To get started, install Pi globally via npm with npm install -g @mariozechner/pi-coding-agent. Configure your preferred AI provider API keys and run pi in your terminal to start an interactive session. Customize your setup by adding AGENTS.md files for project-specific instructions and explore the 50+ example extensions to build your ideal workflow.