An agentic coding CLI for any model

What it does

muyah-code is an open-source coding agent for the terminal. It reads your code, makes the change, runs the tests and tells you straight what happened.

Model choice

Twenty providers, including Anthropic, OpenAI, Gemini, OpenRouter, Groq, DeepSeek and Mistral. Local models through Ollama, LM Studio, llama.cpp, vLLM, or a free Colab GPU.

Safety

It never runs a delete. It shows the exact command and leaves the decision to you. Risky actions like a force-push always ask, whatever the permission mode. /rewind restores code, conversation, or both to an earlier turn.

Working with it

Permission modes (manual, edit, ask, plan, auto) switch with Shift+Tab. It checks its own work with real commands, and /verify runs the checks on demand. /usage breaks down token spend, with daily or per-session budgets. A side panel shows files touched, commands run and sub-agent activity, and past sessions can be replayed.

Extensibility

Sub-agents, skills, MCP servers, hooks and worktrees. It reads Claude Code configuration such as CLAUDE.md. Editor integration with Zed, JetBrains and Neovim through the Agent Client Protocol, and voice input.

Python 3.10 and newer. MIT licence.