Install MundusX Compute

Install the OpenGPU contributor CLI using the production channel in mundusx/releases. These guides describe release 0.2.23. The bootstrap installs the commands; setup selects your connection, contribution cap, workloads and model.

Choose your operating system

• Windows x86_64: PowerShell, managed CUDA/Vulkan or a direct engine.
• Linux x86_64 and ARM64: shell setup, with managed vLLM for supported GB10/GX10 machines.
• Apple Silicon macOS / MLX: native managed MLX or a direct engine. Intel Macs are not included in this release.

Install once using your OS guide

Each platform guide contains the copy-paste command, prerequisites, and recovery steps. Run only the command for your OS. The download page also provides the bootstrap commands.

Windows opens contributor setup automatically. On Linux and macOS, run opengpu install after the bootstrap finishes. For an existing engine, use your OS guide's direct-connection option to skip managed runtime downloads.

Complete setup and connect

1. Choose one inference connection: managed runtime, direct engine, or PAIR validation.
2. Select your contribution cap and workloads. Arrows move, Space toggles workload checkboxes, and Enter continues.
3. Select and prepare a managed model, or verify the exact model on your direct engine.
4. Wait for downloads and verification, then review the setup summary.
5. Run opengpu start --background, opengpu status, and opengpu doctor.
6. Complete onboarding if prompted. Confirm readyForJobs: yes and the expected model before considering setup complete.

Follow the complete setup checklist. Download progress is shown where available; dependency and model-loading steps may show logs instead of a percentage. Setup completion alone does not establish readiness.

Models, media, and external engines

See inference connections for Ollama, LM Studio, vLLM, llama.cpp and standalone models. PAIR is validation-only in this release and cannot start contribution. Keep an external server running with the selected model loaded.

Images and video require at least 32 GiB and 64 GiB respectively after applying the contribution cap, plus compatible hardware and successful local generation verification. Mac media uses MPS/Metal, while MLX serves LLMs.

After reboot or upgrade

Managed models remain saved but must load and warm again at startup. With a direct connection, start that engine and load its selected model before opengpu start --background. Background mode does not itself enable startup at sign-in.

To upgrade, run opengpu update, check opengpu --version, rerun opengpu install, then start and check readiness. Do not delete your identity or models to upgrade.

Contributor access versus Chat API keys

Public contributor installation does not require a Chat API key or GitHub token. Website Chat uses your account login. VS Code and external applications calling a Chat API with key enforcement enabled need a client API key; never use the internal operator token as a client key. Contributor setup and development-agent Chat pairing are separate workflows.