Lemonade Server Snap

A snap package for Lemonade Server - a lightweight, high-performance local AI inference server with an OpenAI-compatible API.

Installation

sudo snap install lemonade-server

Features

• OpenAI-compatible API - Drop-in replacement for OpenAI API endpoints
• Multiple backends - Vulkan, ROCm (AMD GPUs), and CPU support
• Automatic model management - Downloads and caches models from Hugging Face
• Background service - Runs automatically on system startup
• Hardware acceleration - Optimized for AMD GPUs with ROCm support

Supported Hardware

Backend	Hardware	Architecture
Vulkan	Any Vulkan-capable GPU	-
ROCm	AMD RX 7000 series (RDNA3)	gfx110X
ROCm	AMD RX 9000 series (RDNA4)	gfx120X
ROCm	AMD Strix Point APUs	gfx1150
ROCm	AMD Strix Halo APUs	gfx1151
CPU	Any x86_64 processor	-

Quick Start

After installation, the server starts automatically and listens on port 13305.

The snap automatically detects your GPU and selects the best backend:

• AMD GPUs - Uses ROCm for optimal performance
• Other GPUs - Uses Vulkan

GPU Drivers (Vulkan)

Vulkan support is provided by the mesa-2404 content snap, which is installed automatically. If it isn't connected, connect it manually:

sudo snap connect lemonade-server:gpu-2404 mesa-2404:gpu-2404
sudo snap restart lemonade-server.daemon

Check service status

sudo snap services lemonade-server

View logs

sudo snap logs -f lemonade-server.daemon

Test the API

curl http://localhost:13305/api/v1/models

Usage

Load a model

curl -X POST http://localhost:13305/api/v1/load \
  -H "Content-Type: application/json" \
  -d '{"model_name": "Llama-3.2-3B-Instruct-GGUF"}'

Chat completion

curl -X POST http://localhost:13305/api/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "Llama-3.2-3B-Instruct-GGUF",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

List available models

curl http://localhost:13305/api/v1/models

Configuration

Model cache location

Models are cached in /var/snap/lemonade-server/common/.cache/huggingface/.

Using external models

To use models from your home directory or removable media, the snap has access to:

• $HOME (via the home plug)
• /media and /mnt (via the removable-media plug)

Manual server control

# Stop the service
sudo snap stop lemonade-server.daemon

# Start the service
sudo snap start lemonade-server.daemon

# Restart the service
sudo snap restart lemonade-server.daemon

# Disable autostart
sudo snap stop --disable lemonade-server.daemon

Command-line usage

lemonade-server --help

Connecting to Applications

Lemonade Server is compatible with any application that supports the OpenAI API:

• Open WebUI - Set API base URL to http://localhost:13305/api/v1
• AnythingLLM - Configure as OpenAI-compatible endpoint
• Continue (VS Code) - Use OpenAI provider with custom base URL
• LangChain - Use ChatOpenAI with base_url="http://localhost:13305/api/v1"

Troubleshooting

Service won't start

Check the logs for errors:

sudo journalctl -u snap.lemonade-server.daemon.service --no-pager -n 50

GPU not detected

Ensure the gpu-2404 interface is connected:

sudo snap connect lemonade-server:gpu-2404 mesa-2404:gpu-2404
sudo snap restart lemonade-server.daemon

Permission issues

The snap runs in strict confinement. If you need to access files outside the allowed locations, you may need to copy them to your home directory first.

Model download failures

Check your network connection and ensure you have enough disk space:

df -h /var/snap/lemonade-server/common/

Building from Source

git clone https://github.com/lemonade-sdk/lemonade-server.git
cd lemonade-server
snapcraft
sudo snap install --dangerous lemonade-server_*.snap

Links

• Lemonade Server Documentation
• Lemonade SDK GitHub
• Report Issues

License

Snap packaging is GPL-3 License - See LICENSE for details.