Step-by-step for every platform. Windows WSL2 (the most complete guide online), Mac Apple Silicon, Linux, and cloud VPS. One-line install explained, onboarding walkthrough, and common errors with fixes.
NemoClaw's official docs are Linux-first. Windows support works via WSL2, which runs a real Linux kernel inside Windows. This is the most reliable path for Windows users.
Open PowerShell as Administrator and run:
wsl --installThis installs WSL2 with Ubuntu 22.04. Restart your computer when prompted. After reboot, Ubuntu will finish setup and ask you to create a Linux username and password.
Download from docker.com/products/docker-desktop. During setup, ensure "Use WSL2 based engine" is checked. After installation, open Docker Desktop settings → Resources → WSL Integration → Enable integration with your Ubuntu distribution.
Search for "Ubuntu" in the Start menu and open it. You now have a full Linux terminal. Verify Docker works inside WSL2:
docker --versionShould return a version number. If it says "command not found," go back to Docker Desktop settings and enable the WSL2 integration.
In the Ubuntu terminal, run the one-line installer:
curl -fsSL https://www.nvidia.com/nemoclaw.sh | bashThis installs Node.js (if needed), the NemoClaw plugin CLI, and OpenShell. Takes 2 to 5 minutes.
nemoclaw onboardThe wizard asks for your NVIDIA API key, names your assistant, and sets up the initial configuration. See the full wizard walkthrough below.
Colima is a lightweight alternative to Docker Desktop that works well with NemoClaw:
brew install colima
colima start --cpu 4 --memory 8 --disk 40brew install dockercurl -fsSL https://www.nvidia.com/nemoclaw.sh | bashnemoclaw onboardcurl -fsSL https://get.docker.com | sh
sudo usermod -aG docker $USER
newgrp dockerThe last two commands add your user to the docker group so you can run docker without sudo. The newgrp docker applies the group change without requiring a logout.
curl -fsSL https://www.nvidia.com/nemoclaw.sh | bashnemoclaw onboardRunning NemoClaw on a cloud VPS (DigitalOcean, Hetzner, Linode, AWS EC2) is a great way to keep it always on. Use Ubuntu 22.04 LTS as your base image.
For manual VPS setup, follow the Linux steps above. Minimum recommended VPS specs:
The install command is:
curl -fsSL https://www.nvidia.com/nemoclaw.sh | bashBreaking this down:
curl: downloads a file from the internet-fsSL: flags that mean: fail silently on errors, silent progress, follow redirectshttps://www.nvidia.com/nemoclaw.sh: NVIDIA's official install script URL| bash: pipes the downloaded script directly to bash to run itThe script itself:
Run: nemoclaw onboard
The wizard will ask you:
nemoclaw my-assistant connect.
After the wizard completes, your sandbox is ready. The first time it runs, it downloads the ~2.4GB container image, which takes a few minutes depending on your connection.
nemoclaw my-assistant status(Replace "my-assistant" with whatever you named your agent.) Should show the sandbox as running, OpenShell gateway healthy, and inference endpoint connected.
nemoclaw my-assistant connectThis drops you into the sandboxed OpenClaw session. Try sending a message: "What model are you running and what can you do?" If you get a response, everything is working.
nemoclaw my-assistant logsShows real-time logs of what the agent is doing inside the sandbox. Useful for troubleshooting and verifying that security policies are being applied.
Docker isn't running. Fix:
sudo systemctl start docker. If that fails, check sudo systemctl status docker for error details.The 2.4GB sandbox image download is large and can time out on slow connections. Fix: run nemoclaw my-assistant status again. NemoClaw resumes interrupted downloads. If it keeps failing, check your network connection and try again. On a corporate network, ensure Docker has access to ghcr.io (GitHub Container Registry).
The Docker daemon + k3s cluster + OpenShell gateway together require meaningful RAM. If you're on an 8GB machine and have other things running, you may hit memory limits. Fix: close other applications, or configure 8GB of swap space to handle bursts.
# Add 8GB swap on Linux
sudo fallocate -l 8G /swapfile
sudo chmod 600 /swapfile
sudo mkswap /swapfile
sudo swapon /swapfileYour NVIDIA API key isn't configured correctly. Check: cat ~/.nemoclaw/credentials.json. The key should be there. If it's missing or wrong, re-run nemoclaw onboard or manually update the credentials file. Get a new key at build.nvidia.com if needed.
This is expected behavior. The agent tried to reach something not on the default allowlist. You have two options:
~/.nemoclaw/my-assistant/openclaw-sandbox.yaml and add the endpoint to the egress.rules section. Save and the policy reloads without a sandbox restart.Was this guide helpful?