AMD Lemonade is a new open-source local large language model server designed for speed and privacy. It leverages GPUs and NPUs to run efficiently on any PC, offering a unified API for chat, vision, and image generation. The project emphasizes local execution, avoiding cloud dependency for enhanced privacy. Built by the local AI community, it provides a lightweight, native C++ backend under 2MB and supports multiple models simultaneously.

Key features include one-minute installation, auto-configuration for hardware, and compatibility with major frameworks like llama.cpp and Ryzen AI SW. Unified API integration allows apps to utilize chat, speech, and other modalities seamlessly. This development addresses the demand for accessible, on-device AI without compromising performance or openness.