NemoClaw WINDOWS Setup Guide (It Actually WORKS)

Summary

This video walks you through the full process of setting up NemoClaw on Windows, a workaround that most developers have given up on. NemoClaw does not run natively on Windows because it depends on Linux kernel features that Windows simply does not have. The solution is to run it through WSL2 (Windows Subsystem for Linux) with Ubuntu, Docker, and Nvidia GPU passthrough. You start by installing WSL2 and Ubuntu via PowerShell, then enabling systemd inside WSL2 — a step most guides skip that will silently break everything later if you miss it. After that, you install Docker Desktop and configure it to integrate with your Ubuntu WSL2 instance, verified by running a Hello World container test. The most critical and fragile part of the setup is GPU passthrough. You need an up-to-date Nvidia driver installed on the Windows side only — never inside Ubuntu — which automatically creates a stub file that routes GPU calls from Linux back to the real Windows driver. If your GPU does not show up correctly in Ubuntu after this step, you are advised to stop and fall back to the simpler agent-based setup covered in the channel's previous video. Once GPU passthrough is confirmed, you install Node.js (version 20 or above) and npm (version 10 or above) inside Ubuntu, then install the NemoClaw CLI directly from GitHub. The setup wizard walks you through creating a sandbox, entering an Nvidia API key from build.nvidia.com, and selecting a model provider. You can use a paid option like Minimax M2.5 or the free Nvidia Neotron 3 Super 12B model. Finally, you accept default policy presets controlling which external services your agent can access. The honest takeaway: this setup is tedious, error-prone, and involves real repo bugs and version mismatches you will have to troubleshoot live. If GPU passthrough fails at the verification step, do not try to force it — fall back to the agent-based method instead.