Lemonade is the local AI server that gives you the same capabilities as cloud APIs, except 100% free and private. Use the latest models for chat, coding, speech, and image generation on your own NPU and GPU.
Lemonade comes in two flavors:
- Lemonade Server installs a service you can connect to hundreds of great apps using standard OpenAI, Anthropic, and Ollama APIs.
- Embeddable Lemonade is a portable binary you can package into your own application to give it multi-modal local AI that auto-optimizes for your user’s PC.
This project is built by the community for every PC, with optimizations by AMD engineers to get the most from Ryzen AI, Radeon, and Strix Halo PCs.
- Install: Windows · Linux · macOS · Docker · Source
- Get Models: Browse and download with the Model Manager
- Generate: Try models with the built-in interfaces for chat, image gen, speech gen, and more
- Mobile: Take your lemonade to go: iOS · Android · Source
- Connect: Use Lemonade with your favorite apps:
Want your app featured here? Just submit a marketplace PR!
| Platform | Build |
|---|---|
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
To run and chat with Gemma:
lemonade run Gemma-4-E2B-it-GGUF
To code with Lemonade models:
lemonade launch claude
Multi-modality:
# image gen
lemonade run SDXL-Turbo
# speech gen
lemonade run kokoro-v1
# transcription
lemonade run Whisper-Large-v3-Turbo
To see available models and download them:
lemonade list
lemonade pull Gemma-4-E2B-it-GGUF
To manage model aliases for environment-independent naming and active-standby failover:
lemonade alias add production-llm Gemma-4-E2B-it-GGUF
lemonade alias list
# Instant active-standby failover to a different model target
lemonade alias add production-llm Qwen3-0.6B-GGUF
lemonade alias remove production-llm
To see the backends available on your PC:
lemonade backends
For hybrid setups, Lemonade can also route to any OpenAI-compatible cloud provider (Fireworks, OpenAI, OpenRouter, Together, …) alongside local models — see Cloud Offload. (Experimental.)
Lemonade supports a wide variety of LLMs (GGUF, FLM, and ONNX), whisper, stable diffusion, etc. models across CPU, GPU, and NPU.
Use lemonade pull or the built-in Model Manager to download models. Custom GGUF/ONNX models can be pulled from Hugging Face or ModelScope, with their source retained for future updates.
Browse the included model catalog →
Lemonade supports multiple inference engines for LLM, speech, TTS, and image generation, and each has its own backend and hardware requirements.
| Modality | Engine | Backend | Device | OS |
|---|---|---|---|---|
| Text generation | llamacpp |
system |
x86_64/ARM64 CPU, GPU |
Linux |
metal |
Apple Silicon GPU | macOS | ||
cuda |
NVIDIA GPUs (Turing or newer)** | Windows, Linux | ||
vulkan |
x86_64 CPU, AMD iGPU, AMD dGPU; ARM64 CPU/GPU (Linux) |
Windows, Linux | ||
rocm |
AMD GPUs supported by ROCm | Windows, Linux | ||
cpu |
x86_64 CPU; ARM64 CPU (Linux) |
Windows, Linux | ||
nathanw‡ |
AMD Strix Halo (Vulkan performance fork) | Linux | ||
llamacpp-hrx |
hrx† |
AMD GPUs (gfx1100, gfx1151) | Linux | |
flm |
npu |
XDNA2 NPU | Windows, Linux | |
ryzenai-llm |
npu |
XDNA2 NPU | Windows | |
vllm |
rocm† |
Strix Halo iGPU (gfx1151) | Linux | |
ds4 |
rocm‡ |
Prebuilt ds4 for AMD Strix Halo | Linux | |
rocmfpx |
rocm‡ |
AMD Strix Halo | Linux | |
halogen |
rocm‡ |
AMD Strix Halo | Linux | |
| Speech-to-text | whispercpp |
npu |
XDNA2 NPU | Windows |
metal |
Apple Silicon GPU | macOS | ||
vulkan |
x86_64 CPU |
Windows, Linux | ||
rocm |
Supported AMD ROCm iGPU/dGPU families* | Windows, Linux | ||
cpu |
x86_64 CPU |
Windows, Linux | ||
moonshine |
cpu |
x86_64/arm64 CPU |
Windows, Linux, macOS | |
| Text-to-speech | kokoro |
metal |
Apple Silicon GPU | macOS |
cpu |
x86_64 CPU |
Windows, Linux | ||
openmoss |
cuda† |
NVIDIA GPUs | Windows, Linux | |
vulkan† |
Vulkan-capable GPUs | Windows, Linux | ||
| Audio generation | thinksound |
cuda† |
NVIDIA GPUs | Windows, Linux |
vulkan† |
Vulkan-capable GPUs | Windows, Linux | ||
rocm† |
Supported AMD ROCm iGPU/dGPU families (ROCm via TheRock) | Windows, Linux | ||
acestep |
cuda† |
NVIDIA GPUs | Windows, Linux | |
vulkan† |
Vulkan-capable GPUs | Windows, Linux | ||
rocm† |
Supported AMD ROCm iGPU/dGPU families (ROCm via TheRock) | Windows, Linux | ||
| Image generation | sd-cpp |
metal |
Apple Silicon GPU | macOS |
cuda |
NVIDIA GPUs (Turing or newer)** | Windows, Linux | ||
vulkan |
Vulkan-capable GPUs | Windows, Linux | ||
rocm |
Supported AMD ROCm iGPU/dGPU families* | Windows, Linux | ||
cpu |
x86_64 CPU |
Windows, Linux | ||
thenoise |
rocm |
Supported AMD ROCm families | Windows, Linux | |
cuda† |
NVIDIA GPUs | Linux | ||
| 3D generation | trellis |
cuda12† |
NVIDIA Pascal/Volta GPUs (CUDA 12) | Windows, Linux |
cuda† |
NVIDIA GPUs | Windows, Linux | ||
vulkan† |
Vulkan-capable GPUs | Windows, Linux | ||
rocm† |
Supported AMD ROCm iGPU/dGPU families (ROCm via TheRock) | Windows, Linux | ||
| Text classification | onnxruntime |
cpu† |
x86_64 CPU |
Windows |
cpu† |
x86_64/arm64 CPU |
Linux | ||
cpu† |
arm64 CPU |
macOS |
† experimental backend, ‡ guest backend. See Backend Tiers and Formats.
To check exactly which recipes/backends are supported on your own machine, run:
lemonade backends
* See supported AMD ROCm platforms
| Architecture | Platform Support | GPU Models |
|---|---|---|
| gfx1151 (STX Halo) | Windows, Ubuntu | Ryzen AI MAX+ Pro 395 |
| gfx120X (RDNA4) | Windows, Ubuntu | Radeon AI PRO R9700, RX 9070 XT/GRE/9070, RX 9060 XT |
| gfx110X (RDNA3) | Windows, Ubuntu | Radeon PRO W7900/W7800/W7700/V710, RX 7900 XTX/XT/GRE, RX 7800 XT, RX 7700 XT |
** See supported NVIDIA CUDA platforms
| Compute Capability | Architecture | GPU Models |
|---|---|---|
| sm_75 | Turing | RTX 20-series, GTX 16-series, T4 |
| sm_80 / sm_86 | Ampere | RTX 30-series, A100, A40 |
| sm_89 | Ada Lovelace | RTX 40-series, L40, L4 |
| sm_90 | Hopper | H100, H200 |
| sm_100 / sm_120 | Blackwell | RTX 50-series, B100, B200 |
Lemonade's roadmap is defined by a set of working groups. Visit the landing page here to learn each group's goal and roadmap.
Embeddable Lemonade is a binary version of Lemonade that you can bundle into your own app to give it a portable, auto-optimizing, multi-modal local AI stack. This lets users focus on your app, with zero Lemonade installers, branding, or telemetry.
Check out the Embeddable Lemonade guide.
You can use any OpenAI-compatible client library by configuring it to use http://localhost:13305/v1 as the base URL. A table containing official and popular OpenAI clients on different languages is shown below.
Feel free to pick and choose your preferred language.
| Python | C++ | Java | C# | Node.js | Go | Ruby | Rust | PHP |
|---|---|---|---|---|---|---|---|---|
| openai-python | openai-cpp | openai-java | openai-dotnet | openai-node | go-openai | ruby-openai | async-openai | openai-php |
from openai import OpenAI
# Initialize the client to use Lemonade Server
client = OpenAI(
base_url="http://localhost:13305/api/v1",
api_key="lemonade" # required but unused
)
# Create a chat completion
completion = client.chat.completions.create(
model="Gemma-4-E2B-it-GGUF", # or any other available model
messages=[
{"role": "user", "content": "What is the capital of France?"}
]
)
# Print the response
print(completion.choices[0].message.content)Click to learn more about the available APIs and how to embed Lemonade in your own application.
To read our frequently asked questions, see our FAQ Guide
Lemonade is built by the local AI community! If you would like to contribute to this project, please check out our contribution guide.
This is a community project with many maintainers, please see the maintainers list here to see their subject areas. You can reach us by filing an issue or joining our Discord. This project is sponsored by AMD.
Free code signing provided by SignPath.io, certificate by SignPath Foundation.
- Committers and reviewers: Maintainers of this repo
- Approvers: Owners
Privacy policy: This program will not transfer any information to other networked systems unless specifically requested by the user or the person installing or operating it. When the user requests a model download or registry lookup, Lemonade may contact Hugging Face Hub (see their privacy policy) or ModelScope, according to the model source selected by the user or packager.
This project is:
- Built with C++ (server) and React (app) with ❤️ for the open source community,
- Standing on the shoulders of great tools from:
- Licensed under the Apache 2.0 License.
- Portions of the project are licensed as described in LICENSE.