Lemonade Server

Lemonade is a local AI server that provides cloud-API-equivalent capabilities
(chat, coding, speech, image generation) running 100% locally on your own
hardware. It exposes OpenAI, Anthropic, and Ollama-compatible APIs on
localhost:13305, connectable to hundreds of apps.

Features:
- OpenAI, Anthropic, and Ollama-compatible APIs (chat completions, embeddings, etc.)
- Multi-modal inference: LLMs, speech-to-text, text-to-speech, audio generation,
image generation, and 3D generation
- Automatic backend selection based on available hardware — CUDA, ROCm, Vulkan,
NPU, and CPU backends download on demand
- 77+ built-in models (GGUF, FLM, ONNX) with lemonade-server pull / lemonade-server list
for model management
- Cloud offload: route inference to OpenAI-compatible providers alongside
local models
- MCP (Model Context Protocol) client and server support
- Built-in Prometheus metrics endpoint
- Runs as a background service with auto-restart on failure

Supported Hardware:
- AMD GPUs: RDNA2/3/4 (RX 6000/7000/9000), Strix Point/Halo APUs, Instinct MI100/MI200/MI300X/MI350X via ROCm
- AMD XDNA2 NPUs (Ryzen AI) for LLM and speech inference
- NVIDIA GPUs: Turing (sm75) and newer via CUDA (Ampere, Ada Lovelace, Hopper, Blackwell)
- Cross-vendor GPUs via Vulkan (AMD, NVIDIA, Intel, Qualcomm Adreno on ARM64)
- CPU fallback for systems without GPU acceleration
- Both amd64 (x8664) and arm64 (aarch64) architectures

Quick Start:
The server starts automatically after installation. Access the API at: http://localhost:13305/api/v1

Documentation: https://lemonade-server.ai/