Getting Started with Lemonade: Local LLMs in Minutes

Getting Started with Lemonade: Local LLMs in Minutes

In this tutorial, you will deploy Lemonade on your own machine and chat with a local LLM through an OpenAI-compatible API — completely offline. The whole setup takes under ten minutes with Docker.

Step 1 — Pull the image

Lemonade publishes an official container image. Start the server with a single command:

docker run -d --name lemonade -p 8080:8080 \
  -v ./data/lemonade:/data \
  ghcr.io/lemonade-sdk/lemonade:latest

Step 2 — Discover a model

Lemonade includes a model catalog so you can find the right weights for your hardware. The screenshot below shows how the app surfaces models that are optimized for your GPU or NPU:

Step 3 — Download and chat

Pick a model such as Llama-3.2 or Qwen, download it through the interface, and start chatting immediately:

Step 4 — Connect your apps

Because Lemonade speaks the OpenAI protocol, you can point existing tools at http://localhost:8080/v1 without changing your code. It even ships an MCP server, so AI agents can interact with it natively.

Review verdict

Bottom line: Lemonade is the fastest way to go from zero to a private, local AI server. If you value privacy and want zero-cost inference, this is a must-try.