Installation Guide

Get NemoClaw Running

Choose your preferred installation method. All roads lead to the same secure agent stack.

checklist System Requirements

javascript Node.js 20+

v20+ (recommended v22). Auto-installed by the install script via nvm if not present.

computer OS

Ubuntu 22.04+ (recommended). macOS via Colima/Docker Desktop. Windows via WSL2 + Docker Desktop.

deployed_code Container Runtime

Docker (Linux), Colima or Docker Desktop (macOS), Docker Desktop with WSL backend (Windows). Podman on macOS not yet supported.

memory RAM

8 GB minimum, 16 GB recommended

hard_drive Disk

20 GB free minimum, 40 GB recommended

extension GPU

Optional. Required for local Nemotron inference (8GB+ VRAM). Without GPU, inference routes through NVIDIA Cloud API.

One-liner scripts that deploy the complete stack. Recommended for most users. Full guide in docs arrow_forward

Linux / macOS / WSL

Ubuntu 22.04+, macOS (Apple Silicon), Windows WSL2

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

Works on Linux (Ubuntu 22.04+), macOS (Apple Silicon via Colima/Docker Desktop), and Windows (WSL2 + Docker Desktop). The script auto-installs Node.js and runs nemoclaw onboard.

verified Verify Your Installation

Run these commands to confirm NemoClaw is installed correctly.

$ nemoclaw list

Lists all NemoClaw sandboxes. If the command is recognized, the CLI is installed correctly.

$ nemoclaw onboard

Interactive setup wizard: creates a sandbox, configures inference (local Nemotron or NVIDIA Cloud API), and applies security policies.

settings Initial Configuration

After installation, the onboard wizard guides you through setup. Here's what each step does.

Select Your Models

Choose between Nemotron 3 Super 120B (local, full privacy), Nemotron 3 Nano 4B (edge), or cloud models (OpenAI, Anthropic). The Privacy Router handles automatic selection at runtime.

Configure Privacy Policies

Define which data categories stay local and which can be routed to cloud models. Defaults to maximum privacy — everything local.

Set Network Policies

Configure outbound network rules. Default is deny-all. Approve specific domains and APIs that your agent needs to access.

Test Your Sandbox

Run a test agent inside the OpenShell sandbox. Verify isolation, network policies, and model routing are working correctly.

.env — All configuration is stored in ~/.nemoclaw/config.yaml. You can edit this file directly. See the Quickstart guide at /docs/get-started/quickstart for setup details.

system_update_alt Upgrading NemoClaw

Keep NemoClaw up to date to get the latest security patches and features. Upgrade & migration guide arrow_forward

npm

$ npm update -g nemoclaw

pnpm

$ pip install --upgrade nemoclaw

Git

$ cd NemoClaw && git pull && pip install -e .

Docker

$ docker pull ghcr.io/nvidia/nemoclaw:latest && docker restart nemoclaw

Check the changelog at github.com/NVIDIA/NemoClaw/releases for what's new in each version.

delete_forever Uninstalling NemoClaw

If you need to remove NemoClaw from your system, follow the steps for your installation method. Detailed uninstall steps arrow_forward

npm

$ npm uninstall -g nemoclaw

pnpm

$ pip uninstall nemoclaw

Git (Source)

$ rm -rf ~/NemoClaw

Docker

$ docker stop nemoclaw && docker rm nemoclaw && docker rmi ghcr.io/nvidia/nemoclaw:latest

Remove user data (optional)

This deletes all your blueprints, sandbox configs, and model data. Back up anything you need first.

macOS/Linux $ rm -rf ~/.nemoclaw

Windows PS> rmdir /s /q %USERPROFILE%\.nemoclaw

build Troubleshooting

Common issues and how to fix them. More troubleshooting tips arrow_forward

Need Help?

Join the GitHub Discussions for installation support and troubleshooting.

forum Join Discussions menu_book Full Install Docs

hub Ecosystem • article Blog • apps Use Cases