Lemonade Controller 🍋

A cross-platform control center for your Lemonade Server.

Browse and pull models, manage model slots, follow live inference, inspect CPU/GPU/NPU performance, and watch server logs from a responsive desktop or mobile interface.

Download the latest release · All releases · Lemonade Server

Note

Lemonade Controller is an unofficial, community-built companion for Lemonade Server.

What is Lemonade Controller?

Lemonade Server gives you a local API for running AI models. Lemonade Controller puts a full graphical interface on top of it, so you can manage a server on the same machine or elsewhere on your network without reaching for the terminal.

It is built for both quick everyday actions—loading a favourite model or checking a download—and deeper server visibility, including model slots, inference stages, request statistics, hardware telemetry, and live logs.

Screenshots

Features

Model management

• Browse the server's complete model catalog with search, favourites, Custom/Built-in filters, and quantization filters.
• Inspect model metadata, local files, quantization, maximum context window, estimated VRAM usage, and update availability.
• Load, unload, configure, update, or delete models from one interface.
• Configure context size, backends, runtime arguments, image-generation options, model pinning, automatic eviction, idle timeouts, and eviction weighting.
• Save multi-model load presets, estimate their combined VRAM use, reorder entries, and launch the whole setup in one action.

Pulls and downloads

• Paste a Hugging Face repository or lemonade pull command and select from the available quantization variants.
• Automatically fill the suggested model name, recipe, capability labels, and multimodal projection file when the server provides them.
• Follow percentage, current file, bytes transferred, and download speed inline on the Pull page.
• Manage active, paused, completed, and failed jobs from the dedicated Downloads page, including pause, resume, cancel, and remove controls.

Monitoring and observability

• See server health, version, WebSocket port, available model slots, loaded models, pinned models, and models currently loading.
• Track concurrent inference requests live, including auto-loading, prompt, generation, completion, elapsed time, and estimated time remaining.
• Review the latest request's prompt tokens, generated tokens, time to first token, and generation rate.
• Monitor CPU, memory, GPU, VRAM, and NPU utilization with live meters and history charts.
• Stream server logs with severity and tag filters, optional autoscroll, local clearing, and reconnect controls.

Profiles and personalization

• Save and switch between multiple local or remote Lemonade Server profiles.
• Connect through bearer-token authentication, custom HTTP headers, and an optional WebSocket URL override.
• Test a server connection before saving it.
• Choose system, light, or dark appearance and adjust the UI scale.
• Configure model refresh and performance-sampling intervals.
• Export, import, or reset locally stored settings.

Download

Prebuilt packages are published on the GitHub Releases page.

Platform	Available packages
Windows x64	Installer (.exe) or portable archive (.zip)
macOS	Apple Silicon, Intel, or Universal archives (.zip)
Linux x64	Debian (.deb) or RPM (.rpm) package
Android	arm64-v8a, armeabi-v7a, or x86_64 APK
iOS / Web	Build from source

For Android, arm64-v8a is the right choice for most modern phones and tablets. On macOS, choose arm64 for Apple Silicon or x64 for Intel; the Universal build supports both.

Quick start

1. Start a Lemonade Server that is reachable from your device.
2. Download and launch Lemonade Controller.
3. Open Settings → Server Profiles, add the API base URL, and use Test Connection to verify it. The default local URL is http://localhost:8020/api/v1.
4. Open Home for an overview, Models to manage the catalog, or Pull to download a model from Hugging Face.

Lemonade Server 10.10 or newer is recommended for the complete current feature set. The app detects unsupported endpoints and keeps compatible areas available when connected to an older server.

App settings—including profiles, favourites, presets, refresh preferences, appearance, and model parameter overrides—are stored locally on the device.

Build from source

Requirements

• Flutter SDK with a Dart version compatible with ^3.10.8
• A reachable Lemonade Server API endpoint
• Platform build tooling for your chosen target

Run locally

git clone https://github.com/Kidsnd274/lemonade_controller.git
cd lemonade_controller
flutter pub get
flutter run

Build and package

The scripts in scripts/ read the app version from pubspec.yaml and place release artifacts in dist/:

• scripts/build_windows.bat — Windows portable archive and Inno Setup installer
• scripts/build_android.bat — versioned, ABI-split Android APKs
• scripts/build_macos.sh — Apple Silicon, Intel, and Universal archives
• scripts/build_linux.sh — Debian and RPM packages

API coverage

Lemonade Controller uses the configurable Lemonade Server API base URL (http://localhost:8020/api/v1 by default).

Area	Endpoints
Server and hardware	GET /health, GET /system-info, GET /system-stats, GET /stats
Models	GET /models, GET /models/{id}/files
Model lifecycle	POST /load, POST /unload, POST /delete
Pulls	GET /pull/variants, POST /pull
Downloads	GET /downloads, POST /downloads/control
Live activity	Lemonade Server WebSocket log stream

Tech stack

• Flutter and Dart
• Riverpod for application state
• Dio for HTTP
• fl_chart for performance history
• SharedPreferences for local settings
• WebSocket Channel for live server activity

Project structure

lib/
├── main.dart                 # App bootstrap, theme, and UI scaling
├── models/                   # API and local domain models
├── pages/
│   ├── home/                 # Server dashboard and live inference activity
│   ├── models_list/          # Searchable and filterable model catalog
│   ├── model_page/           # Model metadata, files, VRAM, and actions
│   ├── pull/                 # Hugging Face variant picker and pull form
│   ├── downloads/            # Server-managed download queue
│   ├── performance/          # CPU, memory, GPU, VRAM, and NPU charts
│   ├── presets/              # Multi-model load presets
│   ├── logs/                 # Filterable live server logs
│   ├── settings/             # Profiles, appearance, data, and overrides
│   └── widgets/              # Shared responsive navigation
├── providers/                # Riverpod state and async orchestration
├── services/                 # API, WebSocket, and settings services
├── theme/                    # Material theme definitions
└── utils/                    # Formatters, quantization, logging, and VRAM

Feedback

Found a bug or have an idea? Open an issue.