lemonade

Lemonade provides a local inference platform that hosts and serves optimized models (LLMs), image generation, and speech generation directly on your machine. The Lemonade CLI exposes a range of commands to run pre-bundled models, browse and pull models, and manage available backends. Typical workflows include starting the server to access a web UI or API endpoints, listing available models, and pulling specific models for offline use. You can run chat with Gemma or generate images and speech via the built-in interfaces. The CLI supports commands like run to start a specific model, list to view available models, pull to download a model, and recipes to inspect your local backends. This makes it easy to experiment with local, GPU-accelerated inference and integrate Lemonade into other tools and apps.

Prerequisites:

• A supported OS (Linux, Windows, macOS) and a modern Python 3.10–3.13 environment
• Optional: GPU support drivers if you plan to use GPU acceleration

Install steps:

1. Install Python and pip (if not already installed).
2. Create a virtual environment (recommended):
   • python -m venv lemonade-venv
   • source lemonade-venv/bin/activate (Linux/macOS)
   • lemonade-venv\Scripts\activate (Windows)
3. Install the Lemonade server package (example using pipx or pip):
   • Using pip: pip install lemonade-server
   • Or using pipx (isolated env): pipx install lemonade-server
4. Verify installation:
   • lemonade-server --version
5. Run Lemonade with a model (example from README):
   • lemonade-server run Gemma-3-4b-it-GGUF
6. Optional: configure environment variables or supply a config file as needed for your environment.

Tips and notes:

• The Lemonade CLI can pull models locally via lemonade-server pull <model-id> and list models with lemonade-server list.
• For offline or repeatable environments, consider using Model Manager to download and cache models ahead of time.
• If GPU acceleration is used, ensure the appropriate drivers and CUDA toolkit are installed and compatible with your hardware.
• Environment variables (such as LEMONADE_CONFIG) can be used to customize paths, model caches, or backend settings. Review the docs for available options.
• If you encounter port or binding issues, check that no other process is occupying the required port and that firewall rules permit local access.