Download REFEREE

Get started in minutes. Download the installer for your platform and connect any compatible web player to your GPU.

v0.6.0Released April 28, 2026

Pre-release compatibility varies

REFEREE is still pre-release, and feature parity differs across platform and GPU vendor combinations. Review the compatibility chart on the Roadmap page before downloading.

Windows

Download .exe4.8 MB
.exeWindows 10/11

REFEREE.Upscaler_0.6.0_x64-setup.exe · updated April 28, 2026

Installation

  1. 1

    Run the installer

    Double-click the downloaded .exe file. Windows may show a SmartScreen prompt — click “More info” then “Run anyway” to proceed.

  2. 2

    Follow the setup wizard

    Choose your install location and let the installer complete. REFEREE will be added to your Start Menu and system tray.

  3. 3

    Launch REFEREE

    Start REFEREE from the Start Menu or system tray. It will launch a local HTTP server on port 41300 that any compatible web player can connect to.

  4. 4

    Connect your player

    Open a REFEREE-integrated web player. It will automatically detect the local server — no configuration required.

System Requirements

  • Windows 10/11 (64-bit) or modern Linux (x64)
  • NVIDIA GeForce RTX 20-series or newer, or AMD Radeon RX 6000-series or newer
  • Latest NVIDIA or AMD graphics drivers
  • 8 GB+ system RAM

For the terminally inclined

Headless server, Docker, and other install methods

Headless Server

Run only the REFEREE backend pipeline without the desktop GUI. Useful for self-hosted deployments, CI environments, or any machine where you want to expose the upscaling API without the Tauri application.

Rust Binary (from source)

Build and run the headless server binary directly. Configuration is stored at %APPDATA%\com.referee.proxy on Windows and ~/.local/share/com.referee.proxy on Linux.

  1. 1

    Clone the repository

    git clone https://github.com/cmacrowther/referee.git && cd referee
  2. 2

    Build the headless server binary

    Run from desktop/native/

    cargo build --features headless --bin referee-server --release
  3. 3

    Run the server

    ./target/release/referee-server

    The server binds to 0.0.0.0:14002 — reachable from any device on your local network. Set REFEREE_API_TOKEN (≥ 32 chars) to use a known token instead of a generated UUID. Set REFEREE_ALLOWED_ORIGINS to a comma-separated list of origins to pre-approve (e.g. https://myapp.com) so browser clients can obtain the token via POST /v1/auth/request without a desktop UI.

Docker Compose

Pre-built Compose stacks for NVIDIA and AMD. Choose the profile that matches your hardware. The container listens on port 14002 and persists data to a named volume. The NVIDIA profile publishes this on host port 14003 to avoid conflicts with the desktop app; AMD uses 14002 on the host directly.

Prerequisites

NVIDIA

  • nvidia-container-toolkit installed on the host
  • NVIDIA drivers with NVENC support

AMD

  • amdgpu-pro or ROCm driver installed on the host
  • /dev/dri and /dev/kfd device nodes available
  1. 1

    Clone the repository

    git clone https://github.com/cmacrowther/referee.git && cd referee
  2. 2

    Start with your hardware profile

    NVIDIA

    docker compose --profile nvidia up

    AMD

    docker compose --profile amd up

Config files, downloaded encoder binaries, and logs are persisted to the referee_data named volume and survive container restarts.

Authentication environment variables

Add these to the environment: block in your Compose file:

environment:
  REFEREE_API_TOKEN: "your-secret-token-at-least-32-chars"
  REFEREE_ALLOWED_ORIGINS: "https://myapp.com,https://other.com"
  • REFEREE_API_TOKEN — use a known token instead of a generated UUID. Must be ≥ 32 characters. Recommended for production; use Docker secrets or a .env file to avoid committing it.
  • REFEREE_ALLOWED_ORIGINS — comma-separated list of origins pre-approved at startup. Browser clients hitting POST /v1/auth/request from these origins receive the token immediately without needing a desktop UI.
All releases on GitHubIntegration guide