lemonade-sdk/lemonade

Lemonade helps users discover and run local AI apps by serving optimized LLMs right from their own GPUs and NPUs. Join our discord: https://discord.gg/5xXzkMu8Zk

What it does — Lemonade is a program you install on your own computer (Windows, Linux, macOS, Docker) that lets you use modern AI models_without sending data to a cloud service. It covers text chat (LLMs like Gemma and Qwen), speech-to-text (Whisper), text-to-speech, image generation (Stable Diffusion), and even experimental 3D and audio generation.

How it works — You start it once, and it becomes a local "server" that other apps can talk to using standard OpenAI and Anthropic API formats. So instead of paying per request to a cloud, your existing chat apps, coding tools, and custom software can point to localhost and get free, private AI responses. A companion CLI (lemonade run …, lemonade pull …) downloads models and manages them.

Key pieces —

  • Lemonade Server: installs a service; hundreds of apps already integrate with it (Claude Code, AnythingLLM, Dify, GitHub Copilot, Open WebUI, etc.).
  • Embeddable Lemonade: a portable binary you can bundle into your own application to give it local multi-modal AI automatically.
  • Backend matrix: supports llama.cpp (CPU, CUDA, Metal, Vulkan, ROCm), an XDNA2 NPU backend, vLLM and DeepSeek specialized builds for AMD hardware, Whisper.cpp, Kokoro TTS, SD-CPP, and more. It auto-optimizes for the user's PC.
  • Model Manager: browse, download, and update GGUF / FLM / ONNX models from Hugging Face or ModelScope; supports aliases for active-standby failover.
  • Cloud offload (experimental): can route requests to OpenAI-compatible cloud providers alongside local models.

Who it's for — Developers who want a privacy-respecting local AI stack, and especially users of AMD hardware (Ryzen AI / Radeon / Strix Halo), with optimizations contributed by AMD engineers. It is community-built and Apache-licensed, with a roadmap organized into working groups.

Related

  • Project
    momori777/Artemis

    A fully local, private, and uncensored AI companion framework that integrates LLMs, voice synthesis, image generation, and Live2D animations to create interactive, persistent characters.

  • Project
    sauravpanda/BrowserAI

    A JavaScript SDK that enables running production-ready LLMs and multimodal AI models directly in the browser using WebGPU acceleration for privacy and zero server costs.

  • Project
    cubist38/mlx-openai-server

    An OpenAI-compatible API server for running local MLX models on Apple Silicon, supporting text, multimodal, image, and audio models.

  • Project
    ShayneP/local-voice-ai

    A private, low-latency voice assistant that runs locally on your hardware using a stack of speech recognition, LLM, and speech generation models.

  • Project
    mudler/LocalAI

    LocalAI is an open‑source, modular server that lets you run LLMs, vision, speech‑to‑text, text‑to‑speech, image and video models on any hardware. Backends (llama.cpp, vLLM, whisper.cpp, etc.) are pulled as separate OCI images only when needed, providing OpenAI‑compatible APIs, multi‑user auth, privacy‑first on‑prem inference, and built‑in autonomous agents. It works on CPUs, NVIDIA/AMD/Intel GPUs, Apple Silicon, Vulkan and Jetson, and includes distributed scaling and a web UI/CLI.