lemonade

Lemonade is a local AI inference server that runs large language models directly on a user's GPU or NPU, exposing them through OpenAI, Anthropic, and Ollama-compatible REST APIs so that existing tools require no modification to connect. It ships as two variants: Lemonade Server, a background service for Windows, Linux, macOS, and Docker, and Embeddable Lemonade, a portable binary developers can bundle into their own applications. The MCP server component allows Claude Desktop and Claude Code to route requests to locally running models including Llama, Mistral, and Qwen, keeping all inference private and offline. A built-in Model Manager handles browsing and downloading models, while built-in interfaces cover chat, image generation, and speech generation. AMD engineers have contributed specific optimizations for Ryzen AI NPUs, Radeon GPUs, ROCm, and Strix Halo hardware, with Vulkan and ONNX Runtime backends providing the inference layer. Companion iOS and Android apps extend access to the same local server. The project targets developers and privacy-conscious users who want cloud API parity without sending data off-device.

What people ask about lemonade