### Run Interactive Onboarding Setup Wizard

Source: https://docs.nvidia.com/nemoclaw/user-guide/openclaw/reference/commands

The `nemoclaw onboard` command initiates the interactive setup wizard, recommended for new installations. It handles creating an OpenShell gateway, registering inference providers, building the sandbox image, and creating the sandbox. Use this for new installs or recreating a sandbox after configuration changes. Various flags control non-interactive mode, session resume/fresh start, sandbox recreation, GPU usage, Dockerfile source, sandbox naming, agent configuration, control UI port, and acceptance of terms.

```bash
nemoclaw onboard [--non-interactive] [--resume | --fresh] [--recreate-sandbox] [--gpu | --no-gpu] [--from <Dockerfile>] [--name <sandbox>] [--sandbox-gpu | --no-sandbox-gpu] [--sandbox-gpu-device <device>] [--agent <name>] [--agents <agents.yaml>] [--control-ui-port <N>] [--yes | -y] [--no-ollama-autostart] [--yes-i-accept-third-party-software]
```

--------------------------------

### Run NemoClaw Onboarding Wizard

Source: https://docs.nvidia.com/nemoclaw/user-guide/hermes/reference/commands

Launches the interactive setup wizard for new installations. It handles gateway creation, inference provider registration, sandbox image building, and sandbox creation. Use `--non-interactive` for scripted installs.

```bash
nemohermes onboard [--non-interactive] [--resume | --fresh] [--recreate-sandbox] [--gpu | --no-gpu] [--from <Dockerfile>] [--name <sandbox>] [--sandbox-gpu | --no-sandbox-gpu] [--sandbox-gpu-device <device>] [--agent <name>] [--agents <agents.yaml>] [--control-ui-port <N>] [--yes | -y] [--no-ollama-autostart] [--yes-i-accept-third-party-software]
```

```bash
nemohermes onboard [options]
nemoclaw onboard --agent hermes [options]
```

--------------------------------

### Start vLLM Server

Source: https://docs.nvidia.com/nemoclaw/user-guide/hermes/inference/inference-options

Starts a vLLM inference server. This command is an example; any OpenAI-compatible server can be used.

```bash
vllm serve meta-llama/Llama-3.1-8B-Instruct --port 8000
```

--------------------------------

### Non-Interactive Setup for Managed vLLM Installation

Source: https://docs.nvidia.com/nemoclaw/user-guide/hermes/inference/inference-options

Initiate non-interactive setup to install or start a managed vLLM container. This command is sufficient for DGX Spark and DGX Station profiles.

```bash
NEMOCLAW_PROVIDER=install-vllm \
  nemohermes onboard --non-interactive
```

--------------------------------

### Start Fresh Onboarding Session (curl)

Source: https://docs.nvidia.com/nemoclaw/user-guide/hermes/reference/troubleshooting

To start a fresh onboarding session and discard any recorded failed session, pipe the NemoClaw installer script with the `--fresh` argument. This ensures new choices are made.

```bash
curl -fsSL https://www.nvidia.com/nemoclaw.sh | bash -s -- --fresh
```

--------------------------------

### Start Fresh Onboarding Session

Source: https://docs.nvidia.com/nemoclaw/user-guide/openclaw/reference/troubleshooting

If a previous onboarding session failed, you can start fresh by using the `--fresh` flag with the installer script or by setting the `NEMOCLAW_FRESH` environment variable.

```bash
curl -fsSL https://www.nvidia.com/nemoclaw.sh | bash -s -- --fresh
```

```bash
curl -fsSL https://www.nvidia.com/nemoclaw.sh | NEMOCLAW_FRESH=1 bash
```

--------------------------------

### Start Fresh Onboarding Session (env var)

Source: https://docs.nvidia.com/nemoclaw/user-guide/hermes/reference/troubleshooting

Alternatively, start a fresh onboarding session by setting the `NEMOCLAW_FRESH` environment variable to 1 when executing the installer script via bash. This discards the recorded session and provider/model selections.

```bash
curl -fsSL https://www.nvidia.com/nemoclaw.sh | NEMOCLAW_FRESH=1 bash
```

--------------------------------

### Non-Interactive Installation and Onboarding

Source: https://docs.nvidia.com/nemoclaw/user-guide/openclaw/get-started/quickstart

Run both installation and onboarding non-interactively by setting NEMOCLAW_NON_INTERACTIVE and NEMOCLAW_ACCEPT_THIRD_PARTY_SOFTWARE environment variables. This is useful for automated setups.

```bash
curl -fsSL https://www.nvidia.com/nemoclaw.sh | NEMOCLAW_NON_INTERACTIVE=1 NEMOCLAW_ACCEPT_THIRD_PARTY_SOFTWARE=1 bash
```

--------------------------------

### Example Dockerfile for OpenClaw Plugin Installation

Source: https://docs.nvidia.com/nemoclaw/user-guide/openclaw/manage-sandboxes/install-openclaw-plugins

Use this Dockerfile to copy a custom plugin into the OpenClaw extensions directory and ensure OpenClaw refreshes its configuration before NemoClaw starts the sandbox. It handles dependency installation and plugin copying.

```dockerfile
ARG SANDBOX_BASE=ghcr.io/nvidia/nemoclaw/sandbox-base:latest
FROM ${SANDBOX_BASE}

COPY my-plugin/ /opt/my-plugin/
WORKDIR /opt/my-plugin
RUN npm ci --no-audit --no-fund && npm run build

RUN mkdir -p /sandbox/.openclaw/extensions \
 && cp -a /opt/my-plugin /sandbox/.openclaw/extensions/my-plugin \
 && openclaw doctor --fix

WORKDIR /opt/nemoclaw
```

--------------------------------

### Add Package and Model Tooling Presets

Source: https://docs.nvidia.com/nemoclaw/user-guide/openclaw/network-policy/integration-policy-examples

Add presets for installing npm, pypi, brew packages, or accessing Hugging Face models. These are for one-time setup tasks.

```bash
nemoclaw my-assistant policy-add npm --yes
```

```bash
nemoclaw my-assistant policy-add pypi --yes
```

```bash
nemoclaw my-assistant policy-add brew --yes
```

```bash
nemoclaw my-assistant policy-add huggingface --yes
```

--------------------------------

### Export API Key Example

Source: https://docs.nvidia.com/nemoclaw/user-guide/openclaw/get-started/quickstart

Export the API key as an environment variable before launching the installer to avoid being prompted for it during onboarding. Refer to reset instructions if a key is entered incorrectly.

```bash
export NVIDIA_INFERENCE_API_KEY=<your-key>
```

--------------------------------

### Install binutils on Debian/Ubuntu

Source: https://docs.nvidia.com/nemoclaw/user-guide/hermes/get-started/prerequisites

If the installer reports that 'strings' is missing, install 'binutils' using apt-get. This command ensures that the necessary utilities are available for verifying the OpenShell binary.

```bash
sudo apt-get install -y binutils
```

--------------------------------

### Prepare for OpenShell Upgrade During Onboarding

Source: https://docs.nvidia.com/nemoclaw/user-guide/openclaw/reference/commands

For scripted installs where an automatic OpenShell upgrade is desired, set `NEMOCLAW_ACCEPT_EXPERIMENTAL_OPENSHELL_UPGRADE=1` to allow the installer to back up sandbox state, retire the old gateway, install the current OpenShell release, and restore state. If the CLI does not support `backup-all`, manual backup is required before running the installer with `NEMOCLAW_OPENSHELL_UPGRADE_PREPARED=1`.

```bash
curl -fsSL https://www.nvidia.com/nemoclaw.sh | NEMOCLAW_OPENSHELL_UPGRADE_PREPARED=1 bash
```

--------------------------------

### Onboard Model Router with Scripted Setup

Source: https://docs.nvidia.com/nemoclaw/user-guide/hermes/inference/inference-options

Command to onboard the Model Router using environment variables for provider and API key. Use this for scripted or non-interactive setups.

```bash
NEMOCLAW_PROVIDER=routed NVIDIA_INFERENCE_API_KEY=<your-key> nemohermes onboard --non-interactive

```

--------------------------------

### Install and Register Ubuntu 24.04 Distribution

Source: https://docs.nvidia.com/nemoclaw/user-guide/hermes/get-started/prerequisites/windows-preparation

After enabling WSL 2, use this command in an elevated PowerShell to install and register the Ubuntu 24.04 distribution. Ensure the distribution launches to complete first-run setup.

```powershell
wsl --install -d Ubuntu-24.04
```

--------------------------------

### Non-Interactive Installation/Start of Managed vLLM

Source: https://docs.nvidia.com/nemoclaw/user-guide/openclaw/inference/inference-options

Initiate non-interactive onboarding to install or start a managed vLLM container. This is suitable for supported Linux hosts with NVIDIA GPUs. For generic Linux hosts, NEMOCLAW_EXPERIMENTAL=1 is also required.

```bash
NEMOCLAW_PROVIDER=install-vllm \
  nemoclaw onboard --non-interactive
```

--------------------------------

### Prepare for OpenShell Upgrade

Source: https://docs.nvidia.com/nemoclaw/user-guide/hermes/reference/commands

For scripted installs, set `NEMOCLAW_ACCEPT_EXPERIMENTAL_OPENSHELL_UPGRADE=1` to allow the installer to automatically back up sandbox state, retire the old gateway, install the current OpenShell release, and restore state. This is for advanced users managing OpenShell separately.

```bash
NEMOCLAW_ACCEPT_EXPERIMENTAL_OPENSHELL_UPGRADE=1
```

```bash
nemohermes backup-all
```

```bash
openshell gateway remove nemoclaw || openshell gateway destroy -g nemoclaw || openshell gateway destroy
```

```bash
sudo pkill -f openshell-gateway
```

```bash
curl -fsSL https://www.nvidia.com/nemoclaw.sh | NEMOCLAW_OPENSHELL_UPGRADE_PREPARED=1 bash
```

--------------------------------

### Onboard Multiple Sandboxes

Source: https://docs.nvidia.com/nemoclaw/user-guide/hermes/reference/troubleshooting

Demonstrates onboarding multiple sandboxes. The first uses the default port, the second automatically selects the next available port, and subsequent sandboxes can have ports explicitly overridden.

```bash
nemohermes onboard                                                   # first sandbox uses default 18789
```

```bash
nemohermes onboard                                                   # second sandbox uses the next free port
```

```bash
nemohermes onboard --control-ui-port 19000                          # explicit port override
```

--------------------------------

### Resume or Start Fresh Onboarding

Source: https://docs.nvidia.com/nemoclaw/user-guide/hermes/reference/commands

Use `--resume` to continue a previously interrupted onboarding session, preserving recorded settings. Use `--fresh` to discard any saved session state and start the wizard from the beginning.

```bash
NEMOCLAW_SINGLE_SESSION=1 curl -fsSL https://www.nvidia.com/nemoclaw.sh | bash
```

--------------------------------

### Install NemoClaw with Hermes Agent

Source: https://docs.nvidia.com/nemoclaw/user-guide/hermes/get-started/quickstart

Installs the NemoClaw CLI, selects the 'nemohermes' alias, and initiates the guided onboarding flow for the Hermes agent. Ensure Docker is installed and running.

```bash
export NEMOCLAW_AGENT=hermes
curl -fsSL https://www.nvidia.com/nemoclaw.sh | bash
```

--------------------------------

### Non-Interactive Onboarding for Existing vLLM Server

Source: https://docs.nvidia.com/nemoclaw/user-guide/hermes/inference/inference-options

Use this command for non-interactive setup with an already running vLLM server.

```bash
NEMOCLAW_PROVIDER=vllm \
  nemohermes onboard --non-interactive
```

--------------------------------

### Onboard Sandboxes with Distinct Ports

Source: https://docs.nvidia.com/nemoclaw/user-guide/openclaw/reference/troubleshooting

Demonstrates onboarding multiple sandboxes, showing how subsequent sandboxes automatically use the next available port in the default range, or how to explicitly assign a port using `--control-ui-port`.

```bash
nemoclaw onboard
```

```bash
nemoclaw onboard
```

```bash
nemoclaw onboard --control-ui-port 19000
```

--------------------------------

### OpenShell Version Error Message

Source: https://docs.nvidia.com/nemoclaw/user-guide/openclaw/reference/troubleshooting

Example error message when the installed OpenShell version exceeds the maximum supported by the current NemoClaw release.

```text
✗ openshell <version> is above the maximum supported by this NemoClaw release.
  blueprint.yaml max_openshell_version: <max>
```

--------------------------------

### Get Detailed Help for Bootstrap Script

Source: https://docs.nvidia.com/nemoclaw/user-guide/hermes/get-started/prerequisites/windows-preparation

Download the bootstrap script and use Get-Help to view detailed usage information and available parameters for advanced configuration.

```powershell
Get-Help "$env:TEMP\bootstrap-windows.ps1" -Detailed
```

--------------------------------

### Non-Interactive NemoClaw Setup

Source: https://docs.nvidia.com/nemoclaw/user-guide/hermes/get-started/quickstart

Set environment variables for non-interactive installation of NemoClaw. This is useful for CI or scripted environments. Ensure you have your NVIDIA API key.

```bash
export NEMOCLAW_AGENT=hermes
export NEMOCLAW_NON_INTERACTIVE=1
export NEMOCLAW_ACCEPT_THIRD_PARTY_SOFTWARE=1
export NEMOCLAW_SANDBOX_NAME=my-hermes
export NVIDIA_INFERENCE_API_KEY=<your-key>
curl -fsSL https://www.nvidia.com/nemoclaw.sh | bash
```

--------------------------------

### Remove Package Access Presets

Source: https://docs.nvidia.com/nemoclaw/user-guide/openclaw/network-policy/integration-policy-examples

Remove package installation presets like npm, pypi, or brew after a one-time setup task is complete to restrict sandbox access.

```bash
nemoclaw my-assistant policy-remove npm --yes
```

```bash
nemoclaw my-assistant policy-remove pypi --yes
```

```bash
nemoclaw my-assistant policy-remove brew --yes
```

```bash
nemoclaw my-assistant policy-remove huggingface --yes
```

--------------------------------

### Example Build Directory Structure

Source: https://docs.nvidia.com/nemoclaw/user-guide/openclaw/manage-sandboxes/install-openclaw-plugins

This structure shows the typical layout for a Dockerfile and its associated plugin source code. Ensure the Dockerfile and all files it needs to COPY are in the same directory.

```text
my-plugin-sandbox/
├── Dockerfile
└── my-plugin/
    ├── package.json
    └── src/
```

--------------------------------

### Run Windows Bootstrap Script

Source: https://docs.nvidia.com/nemoclaw/user-guide/hermes/get-started/prerequisites/windows-preparation

Execute the PowerShell script to automate the setup of WSL 2, Ubuntu, and Docker Desktop on Windows. This script handles feature enablement, installation, and configuration.

```powershell
Invoke-WebRequest -Uri 'https://raw.githubusercontent.com/NVIDIA/NemoClaw/main/scripts/bootstrap-windows.ps1' -OutFile "$env:TEMP\bootstrap-windows.ps1"; powershell.exe -ExecutionPolicy Bypass -File "$env:TEMP\bootstrap-windows.ps1"
```

--------------------------------

### Accept Third-Party Software for Onboarding

Source: https://docs.nvidia.com/nemoclaw/user-guide/openclaw/reference/commands

This command explicitly accepts the third-party software notice for non-interactive onboarding. It is a prerequisite for certain features like Brave Search.

```bash
nemoclaw onboard --non-interactive --yes-i-accept-third-party-software
```

--------------------------------

### Get Current Inference Provider and Model

Source: https://docs.nvidia.com/nemoclaw/user-guide/hermes/inference/switch-inference-providers

Use this command to display the provider and model the gateway is currently routing to. Run it before switching to confirm the starting state or after a switch to verify the new route.

```bash
nemohermes inference get
```

--------------------------------

### Start Dashboard Port Forward

Source: https://docs.nvidia.com/nemoclaw/user-guide/hermes/manage-sandboxes/lifecycle

Manually restarts a dashboard port forward for a sandbox. Use this if the forward has stopped or the dashboard URL is not loading. Ensure you use the correct port number from your install summary.

```bash
openshell forward start --background <dashboard-port> my-gpt-claw
```

--------------------------------

### Onboard a New Sandbox with a Custom Dockerfile

Source: https://docs.nvidia.com/nemoclaw/user-guide/openclaw/manage-sandboxes/install-openclaw-plugins

Use the `nemoclaw onboard --from` command to point at the Dockerfile in your build directory. This command initiates the sandbox creation process using your custom image.

```bash
nemoclaw onboard --from ./my-plugin-sandbox/Dockerfile
```

--------------------------------

### Deprecated Setup Spark Command

Source: https://docs.nvidia.com/nemoclaw/user-guide/openclaw/reference/commands

The `nemoclaw setup-spark` command is deprecated. Use the standard installer and `nemoclaw onboard` instead, as current OpenShell releases handle older DGX Spark cgroup behavior.

```bash
nemoclaw setup-spark
```

--------------------------------

### Enable Experimental NIM and Onboard

Source: https://docs.nvidia.com/nemoclaw/user-guide/hermes/inference/inference-options

Set the experimental flag to enable NIM features and run the onboard command.

```bash
NEMOCLAW_EXPERIMENTAL=1 nemohermes onboard
```

--------------------------------

### Get Current Inference Provider and Model

Source: https://docs.nvidia.com/nemoclaw/user-guide/openclaw/inference/switch-inference-providers

Use this command to display the provider and model the gateway is currently routing to. Run it before `nemoclaw inference set` to confirm the starting state, or after a switch to verify the new route.

```bash
nemoclaw inference get
```

--------------------------------

### Non-Interactive Onboarding for Anthropic-Compatible Server

Source: https://docs.nvidia.com/nemoclaw/user-guide/hermes/inference/inference-options

Configure non-interactive setup for an Anthropic-compatible server using environment variables.

```bash
NEMOCLAW_PROVIDER=anthropicCompatible \
  NEMOCLAW_ENDPOINT_URL=http://localhost:8080 \
  NEMOCLAW_MODEL=my-model \
  COMPATIBLE_ANTHROPIC_API_KEY=dummy \
  nemohermes onboard --non-interactive
```

--------------------------------

### Install NemoClaw in Ubuntu

Source: https://docs.nvidia.com/nemoclaw/user-guide/hermes/get-started/prerequisites/windows-preparation

After Windows preparation, run this command within the Ubuntu environment to install NemoClaw. It fetches and executes the installation script.

```bash
curl -fsSL https://www.nvidia.com/nemoclaw.sh | bash
```

--------------------------------

### Homebrew Package Installation and Execution

Source: https://docs.nvidia.com/nemoclaw/user-guide/openclaw/network-policy/integration-policy-examples

Add the 'brew' preset, install a Homebrew formula, and execute a formula command within the sandbox. No need to bootstrap Homebrew or source brew shellenv.

```bash
nemoclaw my-assistant policy-add brew --yes
```

```bash
nemoclaw my-assistant exec -- brew install <formula>
```

```bash
nemoclaw my-assistant exec -- bash -lc '<formula-command>'
```

--------------------------------

### Add User to Docker Group on macOS

Source: https://docs.nvidia.com/nemoclaw/user-guide/openclaw/get-started/quickstart

After installation on macOS, if Docker is missing, the installer may prompt for sudo to install it. If your user cannot access the Docker socket, run `newgrp docker` to add your user to the docker group and then re-run the installer.

```bash
newgrp docker
curl -fsSL https://www.nvidia.com/nemoclaw.sh | bash
```

--------------------------------

### Set Ollama Install Mode to System

Source: https://docs.nvidia.com/nemoclaw/user-guide/hermes/reference/commands

Configure `NEMOCLAW_OLLAMA_INSTALL_MODE` to `system` to use the official installer script, which requires sudo, installs to `/usr/local`, and sets up systemd services.

```bash
export NEMOCLAW_OLLAMA_INSTALL_MODE=system
```

--------------------------------

### Rebuild Sandbox with Updated Settings

Source: https://docs.nvidia.com/nemoclaw/user-guide/openclaw/reference/troubleshooting

Exit the sandbox and rerun onboarding to apply most configuration changes. If onboarding failed, use the resume option.

```bash
nemoclaw onboard
nemoclaw onboard --resume
```

--------------------------------

### Onboard with Model Router

Source: https://docs.nvidia.com/nemoclaw/user-guide/openclaw/inference/inference-options

Command to onboard a new system using the routed inference profile, setting the provider and API key.

```bash
NEMOCLAW_PROVIDER=routed NVIDIA_INFERENCE_API_KEY=<your-key> nemoclaw onboard --non-interactive
```

--------------------------------

### Install Specific NemoClaw Release and Check Sandboxes

Source: https://docs.nvidia.com/nemoclaw/user-guide/hermes/manage-sandboxes/lifecycle

Install a specific NemoClaw release by setting the NEMOCLAW_INSTALL_TAG environment variable. After installation, check if existing sandboxes require rebuilding.

```bash
curl -fsSL https://www.nvidia.com/nemoclaw.sh | NEMOCLAW_INSTALL_TAG=v0.0.63 bash
nemohermes upgrade-sandboxes --check
```

--------------------------------

### Configure Slack Channel and Policy

Source: https://docs.nvidia.com/nemoclaw/user-guide/hermes/network-policy/integration-policy-examples

Set Slack bot and app tokens as environment variables, add the Slack channel, rebuild the assistant, and apply the Slack network policy preset.

```bash
export SLACK_BOT_TOKEN=<your-slack-bot-token>
export SLACK_APP_TOKEN=<your-slack-app-token>
NEMOCLAW_NON_INTERACTIVE=1 nemohermes my-assistant channels add slack
nemohermes my-assistant rebuild
nemohermes my-assistant policy-add slack --yes
```

--------------------------------

### Install NemoClaw with a Specific Release Tag

Source: https://docs.nvidia.com/nemoclaw/user-guide/openclaw/manage-sandboxes/lifecycle

Use this command to install or update the NemoClaw CLI, specifying a particular release tag for the installation. This is useful when you need a specific version rather than the latest promoted 'lkg' release.

```bash
curl -fsSL https://www.nvidia.com/nemoclaw.sh | NEMOCLAW_INSTALL_TAG=v0.0.63 bash
```

--------------------------------

### Accept Third-Party Software via Environment Variable

Source: https://docs.nvidia.com/nemoclaw/user-guide/openclaw/reference/commands

An alternative method to accept third-party software for non-interactive onboarding using an environment variable.

```bash
NEMOCLAW_ACCEPT_THIRD_PARTY_SOFTWARE=1 nemoclaw onboard --non-interactive
```

--------------------------------

### Start Cloudflare Named Tunnel

Source: https://docs.nvidia.com/nemoclaw/user-guide/hermes/reference/commands

Starts a Cloudflare named tunnel by setting the CLOUDFLARE_TUNNEL_TOKEN environment variable before executing `nemohermes tunnel start`. The named tunnel hostname and localhost port must be pre-configured in the Cloudflare dashboard.

```bash
export CLOUDFLARE_TUNNEL_TOKEN=<cloudflare-tunnel-token>
nemohermes tunnel start
```