NemoClaw is the channel's recommended self-hosted OpenClaw install — but it doesn't run natively on Windows. The WSL2 workaround exists, and it actually works, but the channel's framing is unusually direct: this is an advanced setup, and "many experienced developers have given up and are waiting for native Windows support." If you want the privacy and control of NemoClaw (vs. a hosted path like KiloClaw or Nut Studio), and you have an Nvidia GPU and the patience to debug WSL2, this is the path. If you want OpenClaw on Windows without the WSL2 pain, go to §10.1 or §10.3 instead.

This article walks through the full WSL2 + Ubuntu + systemD + Docker + Nvidia + NemoClaw install, the failure modes the channel's video explicitly calls out, and the "when to stop and switch to VPS" rule that the channel keeps repeating.

What you'll learn

  • NemoClaw depends on Linux kernel features (Landlock security module, network namespaces, Linux-specific system calls) that the NT kernel does not ship — the WSL2 workaround exists, but it adds complexity at every layer.
  • The install has 8 distinct steps and 1–2 hours of troubleshooting time, minimum. The channel's review is explicit: this is not for everyone.
  • The two failure points the channel flags are WSL2 + systemD (the step almost every guide skips) and the Nvidia driver (you must update the Windows driver to 595.79 from March 10, 2025, and never install a Linux driver inside WSL2).
  • The model side is the same as the VPS NemoClaw path: pick Nvidia Nemotron 3 Super 12B (free) or MiniMax M2.5/M2.7, configure policies (pip, npm by default), and run the OpenClaw TUI.
  • The "stop and switch to VPS" rule is load-bearing. If nvidia-smi doesn't return your GPU after the toolkit install, the channel's review says "stop" — don't loop. The VPS method is genuinely easier.
  • The minimum hardware is Windows 10/11, an Nvidia GPU, 8 GB+ RAM, 20 GB+ free disk, and admin access. Without an Nvidia GPU, the install is unsupported.

Why this is difficult — the Linux kernel dependency

NemoClaw is the channel's recommended self-hosted OpenClaw install, and the reason it's recommended is the security model: Landlock (a Linux-only security module), network namespaces (also Linux-only), and a handful of other kernel features that don't exist on Windows NT. The WSL2 workaround exists because WSL2 ships a real Linux kernel, but the workaround adds complexity at every layer — the channel's review opens with the line "this is an advanced setup with many steps" and does not soften it.

Three things to know before you start:

  1. This is a 1–2 hour minimum install, with troubleshooting on top. The channel's video is 19 minutes long because the video is the summary, not the actual install.
  2. You need an Nvidia GPU. This is the only one of the three Course 10 paths with a hard hardware requirement. Without an Nvidia GPU, GPU passthrough fails and nothing after that works.
  3. Many experienced developers have given up. The channel says this in the video intro, and it's the honest read. The other two Course 10 paths exist for a reason.

The WSL2 path is the right choice if: (a) you specifically need the NemoClaw security model on Windows, (b) you have an Nvidia GPU, (c) you have admin access and time to debug. If any of those is false, use §10.1 or §10.3 instead, or — better — see Course 1 for the VPS path.

The install — 8 steps, 1–2 hours, with the two failure points highlighted

The full install runs through 8 steps. Two of them are the failure points the channel flags; the other six are "follow the instructions, it works."

Step 1: Install WSL2 and Ubuntu (10 minutes)

Open PowerShell as Administrator. Run one command:

wsl --install

This installs WSL2 and Ubuntu in one shot. Restart the PC. On boot, Ubuntu opens and asks for a username (lowercase, no spaces) and a password. The password is the one you'll need every time sudo asks. Don't lose it.

Step 2: Enable systemD (5 minutes, the "step every guide skips")

This is the first failure point. The channel's exact framing: "almost every guide skips this, and it breaks everything later."

Open the Ubuntu terminal. Edit the WSL config:

sudo nano /etc/wsl.conf

If the file is empty, add:

[boot]
systemd=true

If it has content, ensure [boot] systemd=true is present. Save with Ctrl+X, Y, Enter.

Then in PowerShell:

wsl --shutdown

Reopen Ubuntu. Verify:

systemctl status

Should show "running" or "degraded." Both are fine. If it shows "offline," the config didn't save — go back and recheck.

Step 3: Install Docker Desktop (15 minutes)

Download Docker Desktop for Windows AMD64 from docker.com/products/docker-desktop. Install. No need to sign in.

Open Docker Desktop. Settings → Resources → WSL Integration. Enable "Ubuntu-22.04." Click "Apply & Restart."

Verify in Ubuntu:

docker run hello-world

Should print "Hello from Docker!" If it doesn't, the WSL integration didn't apply — go back and re-enable it.

Step 4: GPU passthrough (20 minutes, the make-or-break step)

The second failure point. The channel's framing: "if this fails, nothing after will work."

First, update the Windows Nvidia driver, not the Linux one. The required version is 595.79, released March 10, 2025. Open the Nvidia App, go to System → My Rig, check the driver version, go to Drivers, download 595.79 or newer, install.

Do not install an Nvidia driver inside the Ubuntu WSL2 side. Doing so overwrites the Windows-side stub. The Windows driver installation is what writes the GPU stub into WSL2 — installing a separate Linux driver will conflict with it.

After the Windows driver install, verify in Ubuntu:

nvidia-smi

Should show your GPU model, driver version, CUDA version. Also:

nvcc --version

Should show the CUDA compiler version.

If nvidia-smi returns nothing or errors: stop. The channel's exact instruction is "stop, use VPS method instead, don't waste time in a loop." The supported fallback is the NemoClaw VPS install from §10.2 (or Course 1), not a retry.

Step 5: Install Nvidia Container Toolkit (5 minutes active, 3–5 minutes waiting)

The 4-command sequence:

distribution=$(. /etc/os-release;echo $ID$VERSION_ID)
curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add -
curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | sudo tee /etc/apt/sources.list.d/nvidia-docker.list
sudo apt update
sudo apt install -y nvidia-container-toolkit

Then in Docker Desktop, click "Restart."

Verify again:

nvidia-smi
nvcc --version

Both should work. If nvidia-smi fails here, the channel's review says to stop.

A repo bug 404s the dGPU path; the channel's fix is to swap to Nvidia's generic deb repo manually. The exact fix sequence is in the troubleshooting section below.

Step 6: Install Node.js (3 minutes)

curl -fsSL https://deb.nodesource.com/setup_20.x | sudo -E bash -
sudo apt install -y nodejs
node -v
npm -v

Should show Node v20.x and npm v10.x or higher.

Step 7: Install NemoClaw (15–20 minutes)

npm install -g @nvidia/nemoclaw
nemoclaw init

nemoclaw init runs a setup wizard. When prompted for a sandbox name, pick anything (the channel's example is "boxplant"). Wait for all 22 steps to complete — don't type anything during the install.

When prompted for the Nvidia API key:

  1. Visit build.nvidia.com, create an account (requires 2FA with a phone number), generate an API key.
  2. Paste it into the terminal.

Pick a model provider. The channel's options:

  • Nvidia Nemotron 3 Super 12B — free, 131K context window (smaller than other models, so heavy-context workflows may need to switch).
  • MiniMax M2.5 — if you have an API key. Larger context, more expensive.
  • Other providers as needed.

For the policies step, accept the suggested defaults (Python/pip, Node.js/npm). You can add Discord, Telegram, Slack later.

Step 8: Verify (2 minutes)

Reload the shell:

source ~/.bashrc

Check the sandbox:

nemoclaw sandbox list
nemoclaw sandbox status boxplant

Should show "phase: ready."

Connect:

nemoclaw sandbox connect boxplant
openclaw

The OpenClaw TUI opens. Type:

Hi! How are you?

If you get a response, the install is complete.

The "stop and switch to VPS" rule

The channel's review has one rule that overrides everything else: if nvidia-smi doesn't return your GPU, stop. Don't retry. Don't reinstall. Don't loop. The channel's exact framing is "you will be stuck in a loop otherwise."

The supported fallback is the VPS path from Course 1 — a $2/mo Linux box with NemoClaw installed via the one-line nvidia.com/nemo install. The VPS path:

  • Takes 10–20 minutes, not 1–2 hours
  • Has fewer failure points (no WSL2, no systemD, no Nvidia driver conflict)
  • Is more stable
  • Costs less ($2/mo vs. your electricity bill)
  • Is easier to troubleshoot (no Windows layer)

The channel's framing in Course 5 §5.2 is more direct: "do not run on Windows + Docker Desktop. Hit a lot of problems attempting a Windows local install and abandoned it." The WSL2 path works, but the channel's own preference is the VPS.

The performance reality — 131K context and chunked streaming

Two performance notes the channel flags after install:

  • 131K context window on Nemotron 3 Super. That's smaller than Claude (200K/1M), MiniMax M2.7, or GPT-4. For heavy context workloads, switch to MiniMax M2.7 (larger context) or bring your own API key to a bigger model.
  • Streaming responses may chunk. Don't expect smooth token-by-token streaming. Responses may come in batches depending on the model, network latency, and system resources. This is normal.

The mitigations: pick a model with a larger context window for heavy work, and accept the chunked streaming as the cost of running on Windows.

Common issues and fixes

The channel's review lists six common issues with the WSL2 path. Most are not in scope for this article (they're troubleshooting, not install steps), but the load-bearing ones are:

Issue Diagnosis Fix
WSL2 won't install "WSL 2 requires an update to its kernel component" Download WSL2 kernel update from Microsoft, install, re-run wsl --install
systemctl status shows "offline" systemD not enabled Check /etc/wsl.conf syntax, ensure [boot] systemd=true, run wsl --shutdown
Docker can't find Ubuntu WSL integration not enabled Restart Docker Desktop, check wsl -l -v
nvidia-smi returns "command not found" Driver not installed / not detected Update Windows Nvidia driver to 595.79+; never install a Linux driver inside WSL2
Repository 404 errors on apt update dGPU repo path 404s Manually swap to Nvidia's stable deb repo: sudo rm /etc/apt/sources.list.d/nvidia-docker.list && curl -s -L https://nvidia.github.io/nvidia-docker/stable/ubuntu22.04/nvidia-docker.list | sudo tee /etc/apt/sources.list.d/nvidia-docker.list && sudo apt-key adv --fetch-keys https://nvidia.github.io/nvidia-docker/gpgkey && sudo apt update
NemoClaw sandbox won't start Earlier step failed Check nemoclaw logs; the most common cause is GPU passthrough not working

Try it yourself

The hands-on goal for §10.2 is reproducing the install — and being honest about when to bail.

  1. Verify your hardware first. You need Windows 10/11, an Nvidia GPU, 8 GB+ RAM, 20 GB+ free disk, and admin access. If any of those is missing, skip to §10.1 or §10.3.
  2. Update the Windows Nvidia driver to 595.79 or newer. Do this before you install WSL2. Do not install a separate driver inside WSL2.
  3. Run wsl --install in PowerShell as Administrator. Restart when prompted. Set a username and password you can remember.
  4. Enable systemD. This is the step most guides skip. Edit /etc/wsl.conf, add [boot] systemd=true, save, run wsl --shutdown, reopen Ubuntu, verify with systemctl status.
  5. Install Docker Desktop, enable WSL integration for Ubuntu-22.04, verify with docker run hello-world.
  6. Verify nvidia-smi returns your GPU. This is the make-or-break step. If it returns nothing, stop and switch to the VPS path. The channel is explicit on this.
  7. Install the Nvidia Container Toolkit, restart Docker, verify nvidia-smi and nvcc --version work without errors.
  8. Install Node.js 20.x via NodeSource, then npm install -g @nvidia/nemoclaw.
  9. Run nemoclaw init with a sandbox name, wait for all 22 steps, paste your Nvidia API key, pick a model provider (Nemotron 3 Super is free), accept default policies.
  10. Verify with nemoclaw sandbox status <name> (should show "ready"), nemoclaw sandbox connect <name>, then openclaw. Type "Hi!" and confirm a response.
  11. Time the install. If it took more than 2 hours of active troubleshooting, the channel's review considers that a sign the VPS path would have been faster.
  12. Reconsider the VPS path. A $2/mo NemoClaw VPS is the channel's standing recommendation. The WSL2 path exists for the small minority of users who need self-hosted OpenClaw on Windows specifically. If you're not in that minority, Course 1 is the better path.

Common pitfalls

  • Skipping systemD. "Almost every guide skips this, and it breaks everything later." If you skip the [boot] systemd=true line in /etc/wsl.conf, Docker and Kubernetes will silently fail and you'll spend hours debugging symptoms.
  • Installing an Nvidia driver inside WSL2. The channel's exact instruction is "never." The Windows driver writes the GPU stub into WSL2; a separate Linux driver overwrites it. Update the Windows driver only.
  • Looping on a broken nvidia-smi. If nvidia-smi returns nothing after the toolkit install, stop. Don't reinstall, don't retry, don't tweak. Switch to the VPS path.
  • Using the dGPU repo path. The dGPU repo path 404s. Use the stable deb repo path manually.
  • Picking Nemotron 3 Super for a heavy-context workflow. 131K context is small. For overnight builds, switch to MiniMax M2.7.
  • Blaming the model for chunked streaming. Chunked streaming on WSL2 is normal. If you need smooth streaming, the VPS path will give you a smoother experience.
  • Skipping the WSL2 → VPS comparison. The channel's review is explicit: the WSL2 path works, but the VPS path is faster, cheaper, and easier. Read Course 1 before committing to WSL2.
  • Expecting native Windows support any time soon. The channel's review notes that "native Windows support" is "coming in future updates." Don't bet on it. WSL2 is the path now.
  • Running NemoClaw on a Mac Mini instead. NemoClaw on a Mac Mini is a worse path than NemoClaw on a VPS — see Course 5 §5.3. The Mac Mini's local-network access is a load-bearing risk.
  • Running NemoClaw with your personal Apple ID logged in on the same network. The same "agent on local network" risk that the §5.3 Mac article flags applies here. Keep the Windows box on a guest network or a hotspot if you must run NemoClaw on hardware that shares a network with your personal machines.
  • Reading "advanced setup" as "skip a few steps." "Advanced" means "every step is load-bearing." Skipping the systemD step will break the install two steps later. Don't skim.

Sources

  • NemoClaw WINDOWS Setup Guide (It Actually WORKS) — 6,773 views · video_id: WBZU-LIduto — https://youtu.be/WBZU-LIduto
  • Supabase query — SELECT video_id, title, views, summary_content, summary_key_takeaways FROM public.videos WHERE video_id = 'WBZU-LIduto'; against project ttxdssgydwyurwwnjogq.
  • Cross-references: Course 1: Picking Your Agent Harness (the recommended VPS path), Course 5 §5.2: VPS / cloud (the broader VPS-install framework, including the NemoClaw-on-VPS path the WSL2 path is the Windows equivalent of).