llama.cpp.debugger turns local LLMs into Linux engineers

This project is a significant step toward safe agentic coding by using QEMU as a "blast shield" for local LLM experimentation.

• –Model Context Protocol (MCP) integration allows any compatible client (like Claude Desktop) to grant an LLM full root access to a sandboxed Linux environment.
• –Persistent tmux sessions enable the LLM to maintain GDB state across multiple turns, allowing for complex, iterative debugging that traditional stateless agents struggle with.
• –Optimized for reasoning models like DeepSeek-R1, it separates "thinking" tokens from system commands to improve reliability in technical tasks.
• –Support for GUI applications via virgl and X11 forwarding makes it capable of debugging more than just CLI tools.
• –Automates environment provisioning via cloud-init, reducing the friction for developers to set up a clean, agent-ready "System Under Test."