MCP & Tool-Aufrufe

Lemonade: Local AI Server for GPU and NPU Inference

lemonade-sdk/lemonade

Lemonade is an open-source local AI server that enables users to run optimized Large Language Models (LLMs), speech, and image generation models directly on their own GPUs and NPUs, providing a free and private alternative to cloud APIs.

Offizielles Repository öffnen ↗ Projekt-Startseite ↗
★ 5,2KSterne
⑂ 436Forks
451Offene Issues
C++Sprache
Apache-2.0Lizenz
Q@project.QualityScoreRedaktionelle Bewertung

Projekt-Screenshots

Screenshot von Lemonade: Local AI Server for GPU and NPU Inference Screenshot von Lemonade: Local AI Server for GPU and NPU Inference Screenshot von Lemonade: Local AI Server for GPU and NPU Inference Screenshot von Lemonade: Local AI Server for GPU and NPU Inference Screenshot von Lemonade: Local AI Server for GPU and NPU Inference Screenshot von Lemonade: Local AI Server for GPU and NPU Inference

Übersicht

Lemonade provides a local AI server with capabilities comparable to cloud APIs, focusing on privacy and zero cost. It operates in two modes: Lemonade Server, which installs a service accessible via standard OpenAI, Anthropic, and Ollama APIs by hundreds of applications, and Embeddable Lemonade, a portable binary designed to be packaged directly into third-party applications to provide auto-optimizing multi-modal local AI. The project is built with C++ (server) and React (app), and includes optimizations by AMD engineers for Ryzen AI, Radeon, and Strix Halo hardware.

Wichtige Funktionen

  • Local OpenAI, Anthropic, and Ollama API compatible server
  • Supports text, speech-to-text, text-to-speech, audio, image, and 3D generation
  • Multi-backend support including llamacpp, ONNX Runtime, vLLM, and whispercpp
  • Hardware auto-optimization for AMD Ryzen AI, Radeon, NVIDIA, and Apple Silicon
  • Embeddable portable binary for application integration
  • Built-in Model Manager for browsing and downloading models
  • Command-line interface for model management and inference
  • Cross-platform support for Windows, Linux, and macOS

Voraussetzungen, Installation und Schnellstart

Installation methods vary by platform. Windows users can download an MSI installer; macOS users can download a PKG from the releases page. Linux users can install via Docker, Snap, or distribution-specific packages for Arch, Debian, Fedora, and Ubuntu. Users can also build from source. Detailed installation guides are available in the official documentation.

Nutzung

Use the CLI to run models: `lemonade run Gemma-4-E2B-it-GGUF` for chat, `lemonade run SDXL-Turbo` for image generation, or `lemonade run Whisper-Large-v3-Turbo` for transcription. List available models with `lemonade list`, download them with `lemonade pull <model_name>`, and check supported hardware backends with `lemonade backends`. For API integration, configure OpenAI-compatible clients to use `http://localhost:13305/api/v1` as the base URL.

Modellkompatibilität und Anwendungsfälle

Supports GGUF, FLM, and ONNX model formats. Built-in support includes models like Gemma-4-E2B-it-GGUF, SDXL-Turbo, kokoro-v1, and Whisper-Large-v3-Turbo. Custom GGUF/ONNX models can be pulled directly from Hugging Face or ModelScope.

Lizenz- und Risikohinweise

Licensed under the Apache 2.0 License. Portions of the project may be licensed as described in the repository's LICENSE file. Free code signing is provided by SignPath.io, with certificates by SignPath Foundation.

Editorial verification 2026-08-09: repository URL, owner, description, license and repository statistics were reviewed. License metadata: Apache-2.0. README was fetched for the channel draft; re-check repository dependencies, releases and model terms before production use.

Veröffentlichung und Wartung

Not stated in the repository metadata.

MemPalace

mempalace/mempalace

★ 58KPython

QwenPaw

agentscope-ai/qwenpaw

★ 33,9KPython

Composio

composiohq/composio

★ 29,5KTypeScript