NemoClaw Windows Setup Guide (It Actually Works)

Overview

This is the complete guide to setting up NemoClaw on Windows. NemoClaw does not run natively on Windows, but this workaround using WSL2 (Windows Subsystem for Linux) actually works. Be warned: this is an advanced setup with many steps.

Video Reference: Watch on YouTube
Duration: ~19 minutes
Difficulty: Advanced

⚠️ Important Warning

Before You Start: This setup is complex and error-prone. Many experienced developers have given up and are waiting for native Windows support.

Consider These Alternatives First:

1. Agent-assisted VPS setup (see NemoClaw Setup Guide [blocked]) - Much easier
2. Wait for official Windows support - Coming in future updates
3. Use standard OpenClaw - Simpler and more stable

Only proceed if:

• You're comfortable with advanced troubleshooting
• You have time to debug issues
• You specifically need NemoClaw on Windows
• You have compatible hardware (Nvidia GPU)

Why This Is Difficult

NemoClaw depends on Linux kernel features that don't exist on Windows:

• Landlock security module
• Network namespaces
• Linux-specific system calls

The workaround uses WSL2 to run Linux inside Windows, but this adds complexity and potential failure points.

Prerequisites

Hardware Requirements

• Windows 10/11 (64-bit)
• Nvidia GPU (required for GPU passthrough)
• 8GB+ RAM recommended
• 20GB+ free disk space

Software Requirements

• Administrator access
• Internet connection
• Patience and troubleshooting skills

Installation Steps

Step 1: Install WSL2 and Ubuntu

1.1 Open PowerShell as Administrator

Right-click Start → Windows PowerShell (Admin)

1.2 Install WSL2

This installs WSL2 and Ubuntu in one command.

1.3 Restart Your PC

Required for WSL2 to activate.

1.4 Set Username and Password

After restart, Ubuntu will open automatically. Create:

• Username (lowercase, no spaces)
• Password (you'll need this often)

[Image blocked: Ubuntu Terminal Setup]

Step 2: Enable systemd

⚠️ Critical Step: Almost every guide skips this, and it breaks everything later.

2.1 Open Ubuntu Terminal

Search "Ubuntu" in Start menu.

2.2 Edit WSL Config

Enter your password when prompted.

2.3 Add These Lines

If the file is empty, add:

If the file has content, ensure systemd=true is present.

2.4 Save and Exit

• Press Ctrl+X
• Press Y to confirm
• Press Enter to save

2.5 Shutdown WSL

Open PowerShell and run:

2.6 Reopen Ubuntu and Verify

Should show "running" or "degraded" (both are fine).
If it shows "offline", go back and check /etc/wsl.conf.

[Image blocked: Service Status Check]

Step 3: Install Docker Desktop

3.1 Download Docker

Visit docker.com and download for Windows AMD64.

3.2 Install Docker

Run the installer and follow prompts. No need to sign in.

3.3 Configure WSL Integration

1. Open Docker Desktop
2. Go to Settings → Resources → WSL Integration
3. Enable "Ubuntu-22.04"
4. Click "Apply & Restart"

3.4 Test Docker in Ubuntu

Should show "Hello from Docker!" message.

[Image blocked: Docker Hello World]

Step 4: GPU Passthrough Setup

⚠️ Most Critical Step: If this fails, nothing after will work.

4.1 Update Nvidia Driver (Windows)

1. Open Nvidia App
2. Go to System → My Rig
3. Check driver version
4. Go to Drivers tab
5. Download latest driver (595.79 or newer)
6. Install the driver

Important: Only install on Windows, never inside Ubuntu WSL2.

4.2 Verify GPU in Ubuntu

Open Ubuntu terminal:

Should show:

• Your GPU model
• Driver version
• CUDA version

Should show CUDA compiler version.

⚠️ If GPU Not Detected:

1. Update Windows Nvidia driver
2. Run wsl --shutdown in PowerShell
3. Reopen Ubuntu
4. Try again

If still failing: Stop here. Use VPS method instead. Don't waste time in a loop.

[Image blocked: GPU Detection]

Step 5: Install Nvidia Container Toolkit

5.1 Add Nvidia Repository

5.2 Create Repository List

5.3 Update Package List

5.4 Install Container Toolkit

This takes 3-5 minutes.

5.5 Restart Docker

Go to Docker Desktop and click "Restart".

5.6 Verify Installation

Both should work without errors.

Step 6: Install Node.js

6.1 Add NodeSource Repository

6.2 Install Node.js

6.3 Verify Installation

Should show:

• Node: v20.x or higher
• npm: v10.x or higher

Step 7: Install NemoClaw

7.1 Install NemoClaw CLI

7.2 Run Setup Wizard

7.3 Create Sandbox

When prompted:

• Sandbox name: Choose any name (e.g., "boxplant")
• Wait for all 22 steps to complete (don't type anything)

[Image blocked: NemoClaw Setup Wizard]

7.4 Enter Nvidia API Key

1. Visit build.nvidia.com
2. Create account (requires 2FA with phone)
3. Generate API key
4. Paste into terminal

7.5 Select Model Provider

Choose one:

• Nvidia Nemotron 3 Super 12B (free)
• MiniMax M2.5 (if you have API key)
• Other providers as needed

7.6 Configure Policies

For first setup, accept suggested defaults:

• Python (pip)
• Node.js (npm)

You can add more later (Discord, Telegram, Slack).

Step 8: Verify Installation

8.1 Reload Shell

8.2 Check Sandbox Status

Should show your sandbox (e.g., "boxplant").

Should show phase: "ready".

8.3 Connect to Sandbox

8.4 Open OpenClaw TUI

8.5 Test Your Agent

Type:

If you get a response, congratulations! You've successfully set up NemoClaw on Windows! 🎉

[Image blocked: OpenClaw TUI Success]

Common Issues and Solutions

Issue: WSL2 Won't Install

Error: "WSL 2 requires an update to its kernel component"

Solution:

1. Download WSL2 kernel update from Microsoft
2. Install the update
3. Run wsl --install again

Issue: systemd Not Running

Error: systemctl status shows "offline"

Solution:

1. Check /etc/wsl.conf has correct syntax
2. Ensure systemd=true is under [boot] section
3. Run wsl --shutdown in PowerShell
4. Reopen Ubuntu

Issue: Docker Can't Find Ubuntu

Error: Ubuntu not listed in WSL Integration

Solution:

1. Ensure WSL2 is properly installed
2. Restart Docker Desktop
3. Check that Ubuntu is running: wsl -l -v

Issue: GPU Not Detected

Error: nvidia-smi shows "command not found" or no GPU

Solutions:

1. Update Windows Nvidia driver to latest version
2. Never install Nvidia driver inside Ubuntu
3. Run wsl --shutdown and reopen Ubuntu
4. If still failing, your hardware may not support GPU passthrough

Issue: Repository 404 Errors

Error: "Failed to fetch" or "404 Not Found" during apt update

Solution:

Issue: NemoClaw Sandbox Won't Start

Error: Sandbox creation fails or hangs

Solutions:

1. Check all previous steps completed successfully
2. Verify GPU passthrough works
3. Ensure Docker is running
4. Check logs: nemoclaw logs

Performance Notes

Context Window Limitations

Nvidia Nemotron 3 Super has a 131K context window, which is smaller than other models. For heavy use, consider:

• Switching to MiniMax M2.7 (larger context)
• Using Claude Opus for complex tasks
• Bringing your own API key

Streaming Response

Responses may come in chunks rather than smooth streaming. This is normal and depends on:

• Model choice
• Network latency
• System resources

Next Steps

After successful installation:

1. Test Thoroughly: Try various commands to ensure stability
2. Configure Policies: Add Discord/Telegram if needed
3. Integrate with Existing Agents: Connect to your orchestrator
4. Monitor Performance: Watch for memory/GPU issues
5. Backup Configuration: Save your sandbox settings

Alternative: Use VPS Instead

If you encountered errors during this setup, seriously consider the VPS method:

Advantages:

• 10-20 minute setup (vs 1-2 hours)
• Fewer failure points
• Better stability
• Easier troubleshooting
• Only $2/month

See: NemoClaw Setup Guide [blocked]

Community Support

If you successfully completed this setup:

• Share your experience in Discord
• Help others troubleshoot
• Document any unique issues you solved
• Contribute to improving this guide

If you're stuck:

• Join the community Discord
• Share your error messages
• Ask for help (include hardware specs)
• Consider VPS alternative

Related Resources

• NemoClaw Setup Guide (VPS Method) [blocked]
• Hermes Agent Setup on VPS [blocked]
• OpenClaw Windows Setup [blocked]

Final Thoughts

This setup works, but it's not for everyone. If you made it through, you now have a secure, GPU-accelerated AI agent running on Windows. If you didn't, that's okay—the VPS method is genuinely easier and more reliable.

Last Updated: 2026-05-06