oh-my-pi: Coding Agent Built on Mario Zechner’s Pi

Overview

oh-my-pi (omp) is a terminal-first coding agent built as a coding-focused fork of Pi by Mario Zechner. It aims to give the agent IDE-like capabilities, including language-server operations, debugging, code execution, and structured file editing, rather than relying only on shell commands and text generation.

The project combines a TypeScript agent and CLI with a substantial Rust native core. It supports macOS, Linux, and Windows, and offers interactive terminal, one-shot, SDK, RPC, and editor-facing ACP entry points. Its breadth makes it most relevant to developers who want an extensible, model-flexible agent for real repository work.

Key Features

• IDE integration through language-server operations for diagnostics, navigation, renames, and code actions.
• Debugging through the Debug Adapter Protocol, with support for breakpoints, stepping, and inspecting variables.
• Persistent Python and JavaScript execution, with the ability to call agent tools from the execution environment.
• Parallel subagents, including isolated worktrees and structured results.
• A native tool surface for reading, searching, editing, and working with files, URLs, and repository-related internal paths.
• Multiple interfaces: terminal UI, one-shot prompts, a Node/TypeScript SDK, stdio RPC, and Agent Client Protocol for editors.
• Broad model and provider routing, with configurable roles, custom OpenAI-compatible providers, and fallback chains.
• Cross-platform native components and an embedded shell intended to avoid external-process calls for many common operations.

Use Cases

• A developer wants a terminal-based agent that can use LSP features and make safer, IDE-aware changes across a codebase.
• A team needs an agent to investigate bugs with a real debugger instead of relying only on added print statements.
• A TypeScript application wants to embed an agent session through the project’s SDK or control one over RPC.
• A developer wants to split independent repository tasks among subagents and review their structured findings.
• A user needs one coding-agent workflow across macOS, Linux, and Windows, with configurable model providers.

Getting Started

Install on macOS or Linux with:

curl -fsSL https://omp.sh/install | sh

For Windows, Bun, Homebrew, Nix, pinned versions, and configuration details, see the README.

Alternatives

• openinterpreter: Open Interpreter is Rust-based and targets low-cost models, with support for multiple agent harnesses and ACP-compatible editors.
• tau: Tau is a Python teaching-focused port of Pi, while oh-my-pi adds IDE integration, subagents, and built-in code tools.
• fx: fx is a native Zig agent CLI with an interactive shell and embedding options, rather than an IDE-aware TypeScript fork of Pi.
• zero: Zero is a Go-based agent focused on scriptable assistance, local sessions, and explicit controls over side effects.

Considerations

• The project combines TypeScript with Rust native components, so building from source involves both workspace dependencies and a local native-addon build.
• Its many tools, providers, settings, and integration modes add configuration surface area; teams should review the project’s trust and project-input behavior before enabling it in unfamiliar repositories.
• The repository is actively changing, with frequent commits and releases. That supports momentum but means behavior and interfaces may evolve.
• The measured contributor base is broad rather than concentrated around one person. Recent pull requests are merged quickly, though this does not by itself establish long-term support guarantees.
• The README reports no test-suite status in the supplied repository requirements data, so teams should verify the checks that cover their intended workflows.