Installing OpenClaw + Nemotron 3 Super on
DGXSpark (GB10) and Windows RTX

Notice & Disclaimers

Quick Start Safety Check

Use a clean environment only. Run this demo on a fresh device or VM with no personal data, confidential information, or sensitive credentials. Think of it like a sandbox—keep it isolated.

By installing this demo, you're taking responsibility for all third-party components, including reviewing their licenses, terms, and security posture. Read and accept before you install or use.

What You're Getting

This experience is provided "AS IS" for demonstration purposes only—no warranties, no guarantees. This is a demo, not a production-ready solution. You'll need to implement appropriate security controls for your specific environment and use case.

The demo showcases experimental AI agent capabilities. Even with cutting-edge open-source tools like OpenShell in your toolkit, you need to layer in proper security measures for your specific threat model.

Key Risks with AI Agents

Be mindful of these risks with AI agents:

Data leakage – Any materials the agent accesses could be exposed, leaked, or stolen.
Malicious code execution – The agent or its connected tools could expose your system to malicious code or cyber-attacks.
Unintended actions – The agent might modify or delete files, send messages, or access services without explicit approval.
Prompt injection & manipulation – External inputs or connected content could hijack the agent's behavior in unexpected ways.

Security Best Practices

No system is perfect, but these practices help keep your information and systems safe:

Isolate your environment – Run on a clean PC or isolated virtual machien. Only provision the specific data you want the agent to access.
Never use real accounts – Don't connect personal, confidential, or production accounts. Create dedicated test accounts with minimal permissions.
Never use real accounts – Don't connect personal, confidential, or production accounts. Create dedicated test accounts with minimal permissions.
Vet your skills/plugins – Only enable skills from trusted sources that have been vetted by the community. Review permissions before enabling. During this lab, stick to skills provided or recommended by event staff.
Lock down access – Ensure your OpenClaw UI or messaging channels aren't accessible over the network without proper authentication.
Restrict network access – Where feasible, limit the agent's internet connectivity.
Clean up after yourself – When you're done, remove OpenClaw and revoke all credentials, API keys, and account access you granted.

Lab Participant Acknowledgement

By participating in this demo, you acknowledge and agree that you are solely responsible for your configuration and for any data, accounts, and tools you connect.

To the maximum extent permitted by law, NVIDIA is not responsible for any loss of data, device damage, security incidents, or other harm arising from your configuration or use of demo materials, including OpenClaw or any connected tools or services.

This guide introduces the installation of OpenClaw, powered by Nemotron 3 Super via Ollama, on a DGX Spark (powered by GB10).

Installing OpenClaw + Nemotron 3 Super on DGXSpark (GB10)

Prerequisites

Verify that the OS, GPU, Docker, and Python are available before proceeding with this demo.

head -n 2 /etc/os-release
nvidia-smi
python3 --version

Expected output should show Ubuntu 24.04 (DGX OS), an NVIDIA DGX Spark GB10 GPU, and Python 3.12+.

Step 0: Install OpenClaw

curl -fsSL https://openclaw.ai/install.sh | bash -s -- --no-onboard

Step 1: Install Ollama

Install Ollama (if not already present). Ollama is a local-first platform for inferencing with models.

curl -fsSL https://ollama.ai/install.sh | sh
ollama --version

Verify Ollama is running (it auto-starts as a service after installation). Run this command as follows:

curl http://localhost:11434

Expected output should be: Ollama is running. If not, start it manually:

ollama serve &

Step 2: Download Model

Go to Step 2b if you have a flash drive with pre-downloaded models.

Step 2a: [Online] Pull Nemotron

Pull and run the Nemotron 3 Super model from Ollama. This will take a while—the model is about 86GB.

To test the model via Ollama, run:

ollama pull nemotron-3-super

To load and run the model, please run:

ollama run nemotron-3-super

Step 2b: [Offline] Use local/downloaded models

Use this path when the target hardware has no Internet or you were given a flash drive.

Step 2b.1: Download models (machine with internet)

Skip this if you already have a flash drive with the ./models folder. On a machine with internet access, save the script below as download.sh, then run it. Optionally move the resulting ./models folder to an external disk or flash drive.

#!/bin/bash

models=("qwen3.5:9b" "nemotron-3-super:120b")
for model in "${models[@]}"; do
  dir_name="${model//:/-}"
  ollama pull "$model" && echo "Successfully pulled $model"
  (
    mkdir -p "models/$dir_name" && cd "models/$dir_name" || exit 1
    ollama show "$model" --modelfile > Modelfile
    from_path=$(awk '/^FROM /{print $2; exit}' Modelfile)
    cp "$from_path" model.gguf
    awk '/^FROM / { $0="FROM ./model.gguf" } { print }' Modelfile > Modelfile.tmp && mv Modelfile.tmp Modelfile
    echo "Successfully setup Modelfile for $model"
  ) || exit 1
done

models_line="models=("
for m in "${models[@]}"; do models_line+=" \"$m\""; done
models_line+=" )"

cat > models/setup.sh << EOF
#!/bin/bash
$models_line
for model in "\${models[@]}"; do
  dir_name="\${model//:/-}"
  (cd "\$(dirname "\$0")/\$dir_name" && ollama create "\${dir_name}_local") && echo "Successfully created Ollama model \${dir_name}_local"
done
EOF
chmod +x models/setup.sh
echo "Successfully created setup script for Ollama models"

Step 2b.2: Run the download script

bash download.sh

Step 2b.3: Copy `./models` to target hardware (e.g. DGX Spark)

Copy the ./models folder onto your flash drive, then on the target machine (e.g. DGX Spark) mount the drive and copy the data.

1. Find the drive and partition (e.g. /dev/sdb1 or /dev/nvme1n1p1):

lsblk

2. Create a mount point and mount the partition:

sudo mkdir -p /mnt/ssd
sudo mount /dev/sdb1 /mnt/ssd   # replace sdb1 with your actual partition

3. Copy the models folder to your desired location:

cp -R /mnt/ssd/path/to/models ~/where/you/want/it

4. Unmount the drive when done:

sudo umount /mnt/ssd

Step 2b.4: Prepare models on target hardware

On the target machine, go to the path where you copied the data. The GGUF files and a setup.sh script are in the ./models folder. Run:

cd ./models
bash setup.sh

Step 2b.5: Use the local model

The local model is registered as <model_name>_local. Use that name in the --model flag for all later commands (e.g. nemotron-3-super-120b_local or qwen3.5-9b_local).

Step 3: Launch OpenClaw

To launch OpenClaw, run (use your _local model name if you followed the offline path in Step 2b):

ollama launch openclaw --model nemotron-3-super

The expected output of this should resemble the following screenshot: ollama launch openclaw

Some of the key expected outputs you should see in the logs include:

OpenClaw is running
Open the Web UI:
    http://localhost:18789/#token=ollama

session agent:main:main

Once the OpenClaw agent is ready, you can either enter a prompt on the terminal to interact with the agent, or you can click on the URL (e.g., http://localhost:18789/ …) to open the control UI. From there, you can start messaging the Nemotron 3 Super models, and see all the other features.

This is what the OpenClaw Control UI should look like:

ollama launch openclaw

Step 4: Verify and launch

openclaw models list   # should show nemotron-3-super (or your _local model) as default
openclaw dashboard    # opens a simple dashboard

Messaging Channels + OpenClaw

Once OpenClaw is running, you can connect external messaging channels such as WhatsApp, Telegram, and Discord.

To open the channel configuration workflow, run:

openclaw configure --section channels

You will be prompted to choose which channel you want to set up.

Example: Setting up WhatsApp

In the terminal, select WhatsApp from the list of available channels.
OpenClaw will generate a QR code in the terminal window.
On your phone, open WhatsApp.
Go to Settings > Linked Devices.
Tap Link a Device and scan the QR code shown in the terminal.
After scanning, wait a few moments for the connection to complete.

Once pairing succeeds, OpenClaw will be connected to your WhatsApp account.

Verifying Connection

To verify that WhatsApp is working correctly:

In WhatsApp, open a chat with Yourself.
Send a message to that chat.
OpenClaw should receive the message and respond, allowing you to chat with the agent through WhatsApp.

You can repeat the same openclaw configure --section channels flow later to connect additional platforms such as Telegram or Discord.

Installing OpenClaw + Qwen 3.5 9B on GeForce RTX devices

This guide introduces the installation of OpenClaw, powered by Qwen 3.5 9B via Ollama, on GeForce RTX devices (native Linux or WSL).

Prerequisites

If you are on Windows and WSL is not installed, install it first by running in PowerShell:

wsl --install

Then launch a WSL window:

wsl

Step 1: Install Ollama

Install Ollama (if not already present). Ollama is a local-first platform for inferencing with models.

curl -fsSL https://ollama.ai/install.sh | sh
ollama --version

Expected output should be as follows:

Terminal session: Ollama installation

Verify Ollama is running (it auto-starts as a service after installation):

curl http://localhost:11434

Expected output should be: Ollama is running. If not, start it manually:

ollama serve &

Step 2: Pull Qwen

Next, pull down and run the Qwen model from Ollama.

To test the model via Ollama, run:

ollama pull qwen3.5:9b

To load and run the model, please run:

ollama run qwen3.5:9b

The expected output should be as follows:

Terminal session: Qwen pull and run

Step 3: Launch OpenClaw with Qwen

ollama launch openclaw --model qwen3.5:9b

The expected output should resemble the following screenshot:

Terminal session: Launch OpenClaw with Qwen

Some of the key expected outputs you should see in the logs include:

OpenClaw is running
Open the Web UI:
    http://localhost:18789/#token=ollama

session agent:main:main

Once the OpenClaw agent is ready, you can enter a prompt in the terminal or click the Web UI link to open the control UI and start messaging the model.

This is what the OpenClaw Control UI should look like:

ollama launch openclaw

Step 4: Open the dashboard

Click the Web UI link displayed after a successful launch to access the dashboard.

Messaging Channels + OpenClaw

Once OpenClaw is running, you can connect external messaging channels such as WhatsApp, Telegram, and Discord.

To open the channel configuration workflow, run:

openclaw configure --section channels

You will be prompted to choose which channel you want to set up.

Example: Setting up WhatsApp

In the terminal, select WhatsApp from the list of available channels.
OpenClaw will generate a QR code in the terminal window.
On your phone, open WhatsApp.
Go to Settings > Linked Devices.
Tap Link a Device and scan the QR code shown in the terminal.
After scanning, wait a few moments for the connection to complete.

Once pairing succeeds, OpenClaw will be connected to your WhatsApp account.

Verifying Connection

To verify that WhatsApp is working correctly:

In WhatsApp, open a chat with Yourself.
Send a message to that chat.
OpenClaw should receive the message and respond, allowing you to chat with the agent through WhatsApp.

You can repeat the same openclaw configure --section channels flow later to connect additional platforms such as Telegram or Discord.

Installing OpenClaw + Nemotron 3 Super on NVIDIA Brev

Overview

NVIDIA Brev provides easy access to cloud compute instances across major cloud partners. Developers can select from a broad variety of NVIDIA GPUs and CPUs within a dedicated environment, separate from corporate systems, networks, and data. Use pre-configured Launchables to start building immediately.

For more information, see the NVIDIA Brev documentation.

What You'll Accomplish

Deploy OpenClaw on a Brev cloud instance using a one-click Launchable. The Launchable bootstraps OpenClaw on a fresh Ubuntu-based Brev environment.

No local GPU or manual setup required.

Prerequisites

An NVIDIA Brev account
An NVIDIA Build API key

Step 1: Deploy the OpenClaw Launchable

Click the Launchable link to open the one-click deployment page: Deploy OpenClaw on Brev.
Review the Instance Configuration. The Launchable comes pre-configured with the recommended GPU and environment settings. Adjust if needed.
Click Deploy. Brev will provision a cloud instance, install OpenClaw and code-server, and prepare the environment automatically.
Open the Code Server. Once the instance is running, in the details section of the instance click the "Code Server" button to get started.

Step 2: Running the Demo

On first launch, code-server will automatically open a terminal running configure.sh from the cloned launch-openclaw repo.

Once the gateway is running, you will see output like:

OpenClaw Gateway Started
========================

code-server:
https://code-server0-<id>.brevlab.com

URL:
https://openclaw0-<id>.brevlab.com/chat?session=main

Open the OpenClaw URL in your browser and start chatting with the agent.

Step 3: Messaging Channels + OpenClaw

Once OpenClaw is running, you can connect external messaging channels such as WhatsApp, Telegram, and Discord.

To open the channel configuration workflow, run:

openclaw configure --section channels

You will be prompted to choose which channel you want to set up.

Example: Setting up WhatsApp

In the terminal, select WhatsApp from the list of available channels.
OpenClaw will generate a QR code in the terminal window.
On your phone, open WhatsApp.
Go to Settings > Linked Devices.
Tap Link a Device and scan the QR code shown in the terminal.
After scanning, wait a few moments for the connection to complete.

Once pairing succeeds, OpenClaw will be connected to your WhatsApp account.

Verifying Connection

To verify that WhatsApp is working correctly:

In WhatsApp, open a chat with Yourself.
Send a message to that chat.
OpenClaw should receive the message and respond, allowing you to chat with the agent through WhatsApp.

You can repeat the same openclaw configure --section channels flow later to connect additional platforms such as Telegram or Discord.

Installing OpenClaw + Nemotron 3 Super onDGXSpark (GB10) and Windows RTX

Notice & Disclaimers

What You're Getting

Key Risks with AI Agents

Security Best Practices

Lab Participant Acknowledgement

Installing OpenClaw + Nemotron 3 Super on DGXSpark (GB10)

Prerequisites

Step 0: Install OpenClaw

Step 1: Install Ollama

Step 2: Download Model

Step 2a: [Online] Pull Nemotron

Step 2b: [Offline] Use local/downloaded models

Step 2b.1: Download models (machine with internet)

Step 2b.2: Run the download script

Step 2b.3: Copy ./models to target hardware (e.g. DGX Spark)

Step 2b.4: Prepare models on target hardware

Step 2b.5: Use the local model

Step 3: Launch OpenClaw

Step 4: Verify and launch

Messaging Channels + OpenClaw

Example: Setting up WhatsApp

Verifying Connection

Installing OpenClaw + Qwen 3.5 9B on GeForce RTX devices

Prerequisites

Step 1: Install Ollama

Step 2: Pull Qwen

Step 3: Launch OpenClaw with Qwen

Step 4: Open the dashboard

Messaging Channels + OpenClaw

Example: Setting up WhatsApp

Verifying Connection

Installing OpenClaw + Nemotron 3 Super on NVIDIA Brev

Overview

What You'll Accomplish

Prerequisites

Step 1: Deploy the OpenClaw Launchable

Step 2: Running the Demo

Step 3: Messaging Channels + OpenClaw

Example: Setting up WhatsApp

Verifying Connection

Installing OpenClaw + Nemotron 3 Super on
DGXSpark (GB10) and Windows RTX

Step 2b.3: Copy `./models` to target hardware (e.g. DGX Spark)