Oh My Pi Open source

What it does

Oh My Pi (omp) is a terminal coding agent built by wiring real IDE functionality directly into the agent loop — its tagline is "a coding agent with the IDE wired in." Instead of just editing text, it uses actual Language Server Protocol (LSP) connections for workspace-aware renames/refactors and a real Debug Adapter Protocol (DAP) debugger (lldb, dlv, debugpy) so it can set breakpoints and inspect running programs.

What is inside

• 31 built-in tools including read, write, edit, grep, glob, AST-based structural editing via ast-grep, and archive/PDF handling
• Persistent Python and JavaScript execution environments that survive across tool calls
• A subagent system for parallel task distribution, plus a memory system that retains facts and "lessons learned" between sessions
• Web search across 23 providers with structured extraction, browser automation via Puppeteer, and optional desktop control with screenshots
• Code review with priority-ranked findings, and Git integration for atomic commits and conflict resolution
• Support for 60+ model providers via JSON-RPC / OpenAI-compatible APIs

Works with

omp explicitly imports existing configuration from Claude Code, Cursor, Windsurf, Codex, and Cline, as well as VS Code settings, so switching from one of those tools doesn't mean starting configuration over. It also has native ACP support for the Zed editor and an embedded TypeScript extension system for custom tooling.

How to install or connect

bun install -g @oh-my-pi/pi-coding-agent

Alternatives: curl -fsSL https://omp.sh/install | sh (macOS/Linux), brew install can1357/tap/omp, nix run github:can1357/oh-my-pi, or a PowerShell installer on Windows. Run omp for an interactive session or omp -p "your query" for a single prompt; slash commands like /model, /collab, and /review adjust behavior mid-session.

Maintenance and safety

MIT-licensed and open source, with a large, actively developed codebase (24,000+ commits, ~80,000 lines of Rust at its core plus a TypeScript/Node layer). It performs destructive operations — file writes and deletes — but prompts for permission in interactive mode. Its broader surface (desktop control, code execution, browser automation) means the practical safety of a session depends heavily on which model/provider and permission mode you choose.

Who should use it

Developers who want IDE-grade capabilities (real LSP refactors, real debuggers) inside a terminal agent, and who already use Claude Code, Cursor, Windsurf, Codex, or Cline and don't want to redo their configuration.

Pricing: the agent itself is free and open source (MIT); you bring your own API key or subscription for whichever of the 60+ supported model providers you run it against.