Lemonade

Lemonade is a free, open-source local AI server with a graphical app for downloading, configuring and using models. Desktop and browser clients share the interface; the server owns model files and inference. It also offers a CLI and an embeddable binary for developers.

What it does

• Browse, download and load models through the model manager.
• Chat with compatible local models and connect existing apps through an OpenAI-compatible API.
• Choose supported inference backends for CPU, Vulkan, NVIDIA CUDA, AMD ROCm or Apple Silicon Metal, depending on hardware, model and OS.
• Use supported speech and image models where the selected backend allows it.

The community project is sponsored by AMD, but it is not limited to AMD hardware. Windows 11, Linux and macOS have official installation paths. Check the project's compatibility table before choosing a model or backend; platform support does not mean every accelerator is supported.

Privacy and cost

Local inference runs on your machine. Model downloads and catalog searches contact repositories. Optional cloud routing and telemetry exports can send data elsewhere. Tracing is disabled by default according to the telemetry guide; if enabled, prompts and outputs can be included unless redacted. Local software use needs no subscription, but hardware, electricity and optional cloud services are separate costs.

Lemonade is Apache-2.0 licensed. Individual model licenses are separate.

Screenshot: real Lemonade App model manager from the official documentation. Numbered yellow callouts are part of the guide, not an unmodified app screenshot.

Read our Lemonade discovery. Sources: repository and compatibility table, app guide, telemetry settings. Documentation-based overview, not a hardware benchmark.