can1357/oh-my-pi

What it solves

Oh-my-pi (omp) is a high-capability coding agent designed to bridge the gap between LLMs and the actual development environment. It solves the problem of "shallow" AI coding assistants by deeply integrating with the IDE, debugger, and OS, allowing the agent to perform complex tasks like renaming symbols across a project, debugging segfaults, and managing multi-step workflows without the user having to manually copy-paste code or provide constant context.

How it works

Built on a Rust core, omp provides a comprehensive tool surface that the agent can use to interact with the world. Instead of relying on simple shell commands, it implements many core utilities (like ripgrep and glob) in-process to reduce overhead. It uses a "hashline" editing system that uses content hashes instead of line numbers to ensure edits are precise and avoid corruption. The agent can spawn isolated subagents for parallel work, use a reviewer model to monitor its own progress, and integrate directly with Language Server Protocol (LSP) and Debug Adapter Protocol (DAP) to understand and actually debug code.

Who it’s for

Developers who want a powerful, terminal-based or editor-integrated (e.g., Zed) AI agent that can handle professional-grade software engineering tasks, including debugging, complex refactoring, and cross-platform development (macOS, Linux, Windows).

Highlights

• Deep IDE Integration: Full LSP support for renames and navigation, and DAP support to drive real debuggers (lldb, dlv, debugpy).
• Advanced Editing: "Hashline" patches for token-efficient, precise edits and AST-based structural rewrites.
• Agentic Orchestration: Ability to spawn typed subagents for parallel tasks and a separate "advisor" model for real-time review.
• System-Level Control: In-process coreutils for speed, native desktop control (screenshots, accessibility tree), and a headless browser for web interaction.
• Memory System: Curated memory bank using retain, learn, and recall to maintain codebase knowledge across sessions.
• Collaborative Sessions: Live session sharing via links and QR codes for pair programming with an agent.