Pi for Claude Code Users

Pi is a minimal terminal coding-agent harness. If you already know Claude Code, the core workflow transfers: talk to an agent in your repo, let it read/edit/run commands, and steer it with project instructions.

What transfers immediately

Repo-aware context
Pi loads AGENTS.md and CLAUDE.md from parent directories and the current repo.

Tool-driven coding
Built-in tools cover read, write, edit, and bash.

Slash-command flow
Use /model, /settings, /resume, /compact, /tree, and more.

Session continuity
Sessions auto-save, can be resumed, forked, cloned, exported, or navigated as a tree.

Shortcuts worth learning first

Ctrl+L: open model selector.
Ctrl+P / Shift+Ctrl+P: cycle scoped models.
Shift+Tab: cycle thinking level.
Esc: cancel or abort; press twice to open /tree.
Ctrl+O: collapse or expand tool output.
Ctrl+T: collapse or expand thinking blocks.
Type @ to fuzzy-search project files.
Use !command to run shell output into context; !!command runs without sending output.

Important Pi-specific aspects

Message queue: while Pi works, press Enter to queue steering for the current turn, or Alt+Enter to queue a follow-up after all work finishes.
Branching history: /tree lets you jump to earlier messages and continue from there without losing the original branch.
Customization: Pi is designed to be extended with TypeScript extensions, skills, prompt templates, themes, and pi packages.
Model choice: Pi supports subscription auth and many API-key providers; switch quickly with /model.
Project configuration: use .pi/settings.json, .pi/SYSTEM.md, prompts, skills, extensions, and themes for repo-local behavior.

Mental model: Claude Code gives you a strong default coding agent. Pi gives you a small, hackable harness you can shape into your preferred agent workflow.