Lemonade
A local AI server that runs large language, speech and image-generation models on a user's own GPU or NPU, exposing OpenAI, Anthropic and Ollama-compatible APIs at no cost.
Lemonade comes in two forms: Lemonade Server installs a service that other apps connect to through standard OpenAI, Anthropic or Ollama APIs, and Embeddable Lemonade is a portable binary that can be packaged into another application to add multi-modal local AI that auto-optimizes for the user's PC.
A CLI downloads and runs models for chat, image generation, speech generation and transcription (for example lemonade run Gemma-4-E2B-it-GGUF or lemonade run SDXL-Turbo), lists and pulls available models, and manages named model aliases with active-standby failover between them. It can also route to an OpenAI-compatible cloud provider such as Fireworks, OpenAI, OpenRouter or Together alongside local models as an experimental hybrid mode.
It is written in C++, optimized by AMD engineers for Ryzen AI, Radeon and Strix Halo hardware, installs on Windows, Linux, macOS, Docker or from source, and is licensed under Apache-2.0.