Llauncher serves as a comprehensive management tool for `llama.cpp` `llama-server` instances, designed for seamless interaction by both human operators and LLM agents. It offers a robust MCP (Model Control Protocol) server for programmatic control, enabling agents to list, start, stop, and configure models, alongside a user-friendly Streamlit web UI for direct human oversight. The system includes features for automatic script discovery, persistent configuration storage, and multi-node management across local networks, making it ideal for deploying and orchestrating various LLM models efficiently.
主要功能
01Atomic Just-in-Time (JiT) model swapping with rollback
02Programmatic control via MCP server for LLM agents
03Automatic discovery and persistent configuration of `llama-server` instances
040 GitHub stars
05Web-based Streamlit UI for human operators
06Multi-node management across local networks