PI is an open-source terminal coding agent built around the principle of extensible primitives rather than opinionated workflows. Developers interact with their codebase through a minimal CLI that can read, write, and edit files, run shell commands, and manage multi-file sessions. The agent connects to over 15 LLM providers — including Anthropic, OpenAI, Google, Mistral, and Ollama — with mid-session model switching via keyboard shortcut.
Key capabilities
Minimalist core, extensible by design — PI ships a lean instruction set (under 1,000 tokens) and delegates advanced behaviors to a TypeScript-based Skills system. Teams can add document processing, web search, API integrations, or custom workflows without forking the agent itself.
Multi-provider model support — Connect to Anthropic Claude, OpenAI GPT, Google Gemini, AWS Bedrock, Mistral, Groq, Cerebras, xAI, Ollama, and more from a single install. Switch models mid-session without restarting.
Session branching — Conversations are tree-structured: developers can branch at any point to explore an alternative approach without losing the original thread, then export or share sessions via Hugging Face.
Context engineering tools — Auto-compaction keeps long sessions within context limits; dynamic context injection allows project-specific instructions from AGENTS.md (or CLAUDE.md) to be loaded automatically.
Autonomy level
Level 3 — Supervised agent. PI acts on explicit user instructions and can be interrupted at any point with the Enter key or redirected with a follow-up prompt (Alt+Enter). It does not initiate actions autonomously or run in the background.
Strengths
- No Docker, Python environment, or build steps — pure npm install
- Mid-session model switching without losing context
- Session branching and export make exploratory coding non-destructive
- Skills ecosystem enables custom capabilities without forking the core
- Lightweight footprint suitable for constrained environments (Termux, containers)
Limitations
- No built-in MCP support (requires Skills packages)
- No built-in browser automation or web search (available via Skills)
- No permission system for filesystem/process restrictions — containerization recommended for security boundaries
- No plan mode or structured task management built in
- Terminal-only interface; no web UI