Overview
Lemonade is a local AI server that aims to give you the same capabilities as a cloud API while keeping everything on your own machine. It serves chat, coding, speech and image-generation models from your NPU and GPU, and exposes them through standard OpenAI, Anthropic and Ollama-compatible endpoints so existing clients connect without modification.
It comes in two flavours. Lemonade Server installs as a service you point apps at — the project maintains a marketplace of integrations including Claude Code, AnythingLLM, Open WebUI, OpenHands, Dify, n8n and GitHub Copilot. Embeddable Lemonade is a portable binary you bundle inside your own application, giving it multimodal local AI that auto-optimizes for whatever hardware the user happens to have.
The project is community-built but carries optimizations contributed by AMD engineers targeting Ryzen AI, Radeon and Strix Halo systems, which is why NPU support is unusually first-class here compared to GPU-only local runtimes. Installers ship for Windows, Linux, macOS and Docker, with companion mobile apps for iOS and Android.
What it does
- OpenAI, Anthropic and Ollama-compatible APIs, so existing clients and SDKs work against a local endpoint unchanged
- Runs on NPUs as well as GPUs, with optimizations from AMD engineers for Ryzen AI, Radeon and Strix Halo hardware
- Multimodal out of the box: chat, coding, speech generation and image generation from one server
- Two distributions — a server service, and Embeddable Lemonade as a portable binary you ship inside your own app
- A built-in Model Manager for browsing and downloading models, plus built-in web interfaces for each modality
- A marketplace of verified app integrations including Claude Code, Open WebUI, AnythingLLM, OpenHands, Dify and n8n
Getting started
Lemonade ships installers rather than a package-manager one-liner. Install the server, pull a model with the Model Manager, then point any OpenAI-compatible client at the local endpoint.
Install the server
Windows has an MSI installer; Linux, macOS, Docker and from-source paths are documented in the install guide at lemonade-server.ai. A Docker image is available if you would rather not install a service.
Get models
Browse and download models from the built-in Model Manager, which handles picking a build optimized for your NPU or GPU.
Generate
Try the model straight away in the bundled interfaces for chat, image generation and speech generation before wiring anything up.
Connect your apps
Point any OpenAI-, Anthropic- or Ollama-compatible client at the local server. The marketplace lists verified setups for Claude Code, Open WebUI, AnythingLLM, OpenHands, Dify, n8n and GitHub Copilot.
Commands and code are distilled from the project's own documentation — always check the official repo for the latest.
When to use it
- Run a coding assistant entirely on a laptop NPU so no source code leaves the machine
- Give an existing OpenAI-compatible app a local backend by changing only its base URL
- Ship a desktop application with offline multimodal AI built in, using Embeddable Lemonade rather than requiring a cloud key
- Get real use out of the NPU in a Ryzen AI or Strix Halo machine, which most local runtimes ignore in favour of the GPU
How Lemonade compares
Lemonade alongside other open-source local runtimes tools AI/TLDR tracks, ranked by GitHub stars.
| Tool | Stars | What it does |
|---|---|---|
| Ollama | ★ 181k | A developer-friendly tool that downloads and runs local LLMs from the terminal with a built-in OpenAI-compatible API. |
| llama.cpp | ★ 128k | A C/C++ inference engine that runs LLMs in the GGUF format on CPUs, Apple Silicon, and GPUs with low memory use. |
| GPT4All | ★ 77.4k | GPT4All is a free desktop app and Python client that runs large language models locally on your own computer, with no API calls or GPU required. |
| LocalAI | ★ 49.1k | A self-hosted server that exposes an OpenAI-compatible API for running text, vision, voice, and image models on local hardware. |
| Jan | ★ 44.5k | An open-source desktop app that runs LLMs fully offline as a ChatGPT-style assistant on your own computer. |
| llmfit | ★ 36.7k | A Rust terminal tool that inspects your CPU, RAM, GPUs and VRAM and scores which open-weight models and quantizations will actually run well on that machine, with a TUI, CLI, REST API and local-runtime integrations. |
| Colibrì | ★ 35k | A pure-C inference engine that keeps a Mixture-of-Experts model's dense trunk resident in RAM and streams its routed experts from disk, so 744B-2.8T models run on consumer hardware. |
| Lemonade | ★ 5.7k | A local AI server that serves optimized models from your own NPU and GPU |