NemoClaw

Deploy NemoClaw, an open source CLI orchestrator from NVIDIA that runs the OpenClaw AI agent inside a Docker sandbox and proxies its network access through OpenShell, a companion process.

You can configure and manage the system using a CLI onboarding wizard that selects an inference provider, collects credentials, and starts the sandbox dashboard. This app creates a limited user on the system named nemoclaw and configures system services for automatic sandbox management.

1. Deploy NemoClaw

 Estimated deployment time: 10–15 minutes

Software included

Software installed on your Linode when deploying the app.

Base distribution and plan

• Supported distribution: Ubuntu 24.04 LTS
• Recommended plans: Use at least a Linode 8GB or higher plan to comfortably run the OpenClaw agent and related services.

Configuration options

StackScript ID: 2164119

Notes:

• After the app deployment completes, the password for your limited sudo user is generated and stored in the .credentials file in the home directory, along with application-specific passwords. Log in to your instance as root through the Lish console or SSH, then run cat /home/$USERNAME/.credentials to view its contents.
• You can automatically configure a custom domain (optional) during deployment. To do that:
  1. First, configure your domain to use Linode’s name servers via your registrar. See Configure your domain's authoritative name servers for details.
  2. Once you complete that, specify the Linode API token, subdomain, and domain for the app in Cloud Manager or via API/CLI/Terraform.

Use API, CLI, or Terraform

curl --location 'https://api.linode.com/v4/linode/instances' \
    --header 'Content-Type: application/json' \
    --header 'Accept: application/json' \
    --header 'Authorization: Bearer abc123def456hij789klm' \
    --data-raw '{
        "region": "us-east",
        "type": "g6-standard-2",
        "image": "linode/ubuntu24.04",
        "label": "my-nemoclaw-one-click-app",
        "root_pass": "@C0mpl3x#P@ssw0rd",
        "stackscript_id": 2164119,
        "stackscript_data": {
            "user_name":"jsmith",
            "soa_email_address":"jsmith@example.com"
        }
    }'

linode-cli linodes create \
    --region us-east \
    --type g6-standard-2 \
    --label my-nemoclaw-one-click-app \
    --image linode/ubuntu24.04 \
    --root_pass @C0mpl3x#P@ssw0rd \
    --stackscript_id 2164119 \
    --stackscript_data '{"user_name":"jsmith", "soa_email_address":"jsmith@example.com"}'

resource "linode_instance" "my-linode" {
    region         = "us-east"
    type           = "g6-standard-2"
    label          = "my-nemoclaw-one-click-app"
    image          = "linode/ubuntu24.04"
    root_pass      = "@C0mpl3x#P@ssw0rd"
    stackscript_id = 2164119
    stackscript_data = {
        "user_name"         = "jsmith"
        "soa_email_address" = "jsmith@example.com"
    }
}

2. Access NemoClaw

Onboard NemoClaw

Once the deployment is complete, go through the onboarding wizard to configure your inference provider and start the dashboard. This app triggers onboarding for you when you log in as root.

1. Log in to your instance.

   If you disabled root login to the server during the setup of the NemoClaw app, log in as a sudo user.

   Bash

   ssh root@YOUR_INSTANCE_IP

   Replace YOUR_INSTANCE_IP with the IP address of your Linode instance.

2. Escalate privileges to root. Once you’ve logged in, the motd (message of the day) appears.

   Bash

   *********************************************************
   Akamai Connected Cloud NemoClaw Quick Deploy App
   Dashboard Access: SSH tunnel required (see details below)
   Credentials File: /home/admin/.credentials
   Documentation: https://techdocs.akamai.com/quick-deploy-apps/docs/nemoclaw
   *********************************************************

3. Become root by running sudo su -. When prompted for the sudo password, paste it from the .credentials file.

4. The onboarding script runs automatically, and you're prompted to start the setup wizard for NemoClaw.

   Bash

   Do you want to run the nemoclaw onboard wizard? [y/n]:

5. If you're ready to perform the onboarding, enter y. It takes you to the nemoclaw onboard wizard where you can complete the configuration. It prompts you to:

   • Select an inference provider: Choose from the supported options, including NVIDIA Build/Endpoints, OpenAI, Anthropic, OpenRouter, Gemini, or a self-hosted OpenAI-compatible server.

   • Supply provider credentials: Enter the API key or connection details for your chosen provider.

   Important: NemoClaw requires an external Large Language Model (LLM) to function. You must provide valid credentials for at least one of the supported inference providers during onboarding.

   Once the setup is complete, the onboarding script is removed, and it won’t prompt again on the next login.

Access the dashboard

The NemoClaw dashboard isn't exposed on a public HTTP(S) endpoint. Instead, access it securely through SSH tunneling from your local machine.

1. On your local machine, configure an SSH tunnel.

   Command

   ssh -L 18789:127.0.0.1:18789 root@YOUR_INSTANCE_IP

   Replace YOUR_INSTANCE_IP with your Linode instance’s IP address. This forwards the port 18789 on your local machine to the dashboard port on the instance.

2. While the tunnel is open, get the dashboard URL and token by running this command on the instance.

   CommandOutput

   sudo -i -u nemoclaw nemoclaw dashboard-url

   Dashboard URL:
   http://127.0.0.1:18789/#token=1Ab23cd45eFgh67iJk8lmN9012oP3qR
   Treat this URL like a password -- do not log, share, or commit it.

3. Open your browser and visit the local URL with the token to access the dashboard.

   Important: The dashboard token is sensitive. Treat it like a password and don't share or commit it to version control.

Additional resources

For more information about the installed packages, see their official documentation.

• NemoClaw GitHub Repository for the latest updates and community support.
• NemoClaw Documentation for advanced features and configurations.
• Inference Provider Documentation for optimizing your chosen AI model endpoint.