The Open Source Coding Agent
Bring AI into your dev workflow — transparent, fast, and fully in your control.
Demo: Muonry in action.
Built in <1200 Lines
Why we created Muonry — a transparent, hackable coding agent
The Problem
Existing AI coding tools are black boxes. You can't see what they're doing, can't modify their behavior, and can't run them locally. They're expensive, opaque, and locked behind APIs.
Our Solution
Muonry is an open, free agent that anyone can understand, modify, and build upon. We've kept the entire codebase under 1200 lines of code — small enough to read in one sitting, powerful enough to be genuinely useful.
Fast & Lightweight
Built for low latency and minimal resources — written in Rust, with a tiny surface area you can reason about.
Fully Transparent
100% open source. No hidden APIs.
Extensible
Bring your own provider under the hood — Groq, Cerebras, OpenRouter, and more. Works with Python, JS, and beyond.
Interactive Demo
Try refactoring code on the left; Muonry explains and improves on the right.
You ask for a change (e.g., “Refactor function X”). Muonry plans the task and applies it to your repo using safe, auditable tool calls:
read_file/grep: find code and gather context.apply_patch: make precise edits via small diffs.run_shell/smart_run_shell: run builds/tests; suggest or apply safe fixes.update_plan: keep a short, evolving TODO of steps.interactive_shell: only for short CLI wizards (e.g., scaffolding).
Everything happens locally in your workspace. You can use any LLM provider by setting API keys (Groq, Cerebras, OpenRouter, etc.). Secrets aren’t printed, and destructive actions are avoided unless you ask for them.
Architecture Overview
Terminal/CLI → Muonry Core → Tools → Providers → Project Files
Community
Muonry is open to everyone — start coding with your AI partner today.