AtlasComputeNetworkDevelopersSign inGet started
MundusX Documentation

Windows installation

OpenGPU contributor setup for release 0.2.23.

Windows contributor installation

This guide installs an OpenGPU compute contributor on Windows x86_64. It uses the public release in mundusx/releases. No Rust, GitHub token, Chat pairing, or development workspace is required for this contributor-only path.

1. Prepare Windows

Use an ordinary PowerShell window under the account that will contribute. Have a working GPU driver, internet access, and enough disk space for the runtime and chosen models. NVIDIA machines use the CUDA runtime by default; other machines receive Vulkan. Runtime download alone does not prove GPU support.

Do not paste the Linux/macOS curl ... | bash command into PowerShell.

2. Download and run the installer

Paste this entire line into PowerShell:

$installerPath = Join-Path $env:TEMP "mundusx-install.ps1"; Invoke-WebRequest -Uri "https://github.com/mundusx/releases/releases/download/opengpu-prod/install.ps1" -OutFile $installerPath; powershell -NoProfile -ExecutionPolicy Bypass -File $installerPath -SetupMode contributor -SkipTrayAutoStart

This explicitly selects compute contribution and disables tray startup at sign-in. Without -SetupMode contributor, the installer offers developer, contributor, or both. Without -SkipTrayAutoStart, it configures tray auto-start.

Already using a local engine? To skip the managed CUDA/Vulkan download, use this command instead of the line above (replace the example URL with the actual engine endpoint):

$installerPath = Join-Path $env:TEMP "mundusx-install.ps1"; Invoke-WebRequest -Uri "https://github.com/mundusx/releases/releases/download/opengpu-prod/install.ps1" -OutFile $installerPath; powershell -NoProfile -ExecutionPolicy Bypass -File $installerPath -SetupMode contributor -SkipTrayAutoStart -Connection direct -ClusterUrl http://127.0.0.1:1234

Start Ollama, LM Studio, vLLM, or llama.cpp before setup. LM Studio must have an LLM loaded and its API server enabled. The normal installer line downloads the managed GPU runtime before the CLI wizard, even if you later choose a direct engine there; use -Connection direct to avoid that download.

Expect these six installer phases, with download progress where available:

  1. Downloading OpenGPU CLI.
  2. Verifying the signed release.
  3. Downloading the GPU runtime, or skipping it for an external connection.
  4. Downloading the OpenGPU node agent.
  5. Downloading the Windows tray application.
  6. Installing verified components.

The default binary directory is %USERPROFILE%\.opengpu\bin. A fresh PowerShell window opens with opengpu install running. If it does not, run:

& "$env:USERPROFILE\.opengpu\bin\opengpu.exe" install

3. Complete the setup wizard

Follow the shared setup choices: choose one connection (managed model, direct engine, or PAIR validation), then complete the contribution cap, workloads, and any managed model/runtime preparation. Direct setup verifies the exact selected model. PAIR currently validates only and cannot contribute. Review the setup summary. Wait for model downloads to finish. Public contribution does not ask you to paste a GitHub token or account password.

4. Start and verify

In PowerShell:

$opengpu = "$env:USERPROFILE\.opengpu\bin\opengpu.exe"
& $opengpu start --background
& $opengpu status
& $opengpu doctor

& $opengpu --version should report opengpu 0.2.23 for this release.

Complete contributor onboarding if prompted. Look for readyForJobs: yes and check the reported connection/model state. A completed download or saved setup is not the same as a connected, ready contributor. Follow any reason shown by status or doctor before proceeding.

5. Everyday commands

& $opengpu pause
& $opengpu resume
& $opengpu cap
& $opengpu model list
& $opengpu exit

The $opengpu variable lasts only in the current PowerShell window; set it again in a new window. After reboot, start an external engine and load its selected model first if using a direct connection, then run start --background. For installation errors, inspect %USERPROFILE%\.opengpu\logs\installer.log. For node diagnostics, run & $opengpu logs. Redact secrets before sharing logs.

For optional images/video, see media setup. Managed Windows ComfyUI currently requires NVIDIA CUDA and Python 3.12 or 3.13; a successful Vulkan LLM installation does not imply managed media support.