NVIDIA NemoClaw

NVIDIA's open stack for running OpenClaw, Hermes and LangChain Deep Agents in OpenShell sandboxes with network policy and managed inference

CLIEditors’ pickHigh risk

nvidia/nemoclaw

Install

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

Command from NVIDIA's quickstart docs. Check the prerequisites first.

This is third-party code. Review the repository files before installing.

What it does

NemoClaw places the chosen agent inside an NVIDIA OpenShell sandbox and manages it through the nemoclaw CLI. The installer walks through onboarding: agent, sandbox name, inference provider and model, with an express mode on supported DGX and WSL hosts. The CLI then shows status, prints the dashboard URL, launches the agent TUI and connects to the sandbox. The stack also handles network policy, integrations, snapshots and sandbox lifecycle.

Who it is for. For teams and enthusiasts who want to run autonomous agents like OpenClaw with isolation and network control.

Good fit when

  • You want to run OpenClaw or Hermes without letting it into your main system
  • You need network policy and managed credentials for the agent
  • You have a DGX or WSL machine and want a guided installer

Not a fit when

  • You need a stable product with guaranteed support: the project is alpha
  • Your agent is not on the supported list

Example request

Help me install NemoClaw with OpenClaw following the official docs, asking before each command

Limitations

Alpha; maintainers respond on a best-effort basis. Only OpenClaw, Hermes and LangChain Deep Agents Code are supported. The installer downloads third-party software including Node.js and requires accepting its terms. System requirements are in NVIDIA's docs.

How to disable. Stop and remove sandboxes with the nemoclaw CLI following the uninstall section of NVIDIA's docs, then remove the CLI.

Security check

  • A network installer sets up system software
  • Manages autonomous agents and inference credentials

README in short

The README presents NemoClaw as a reference stack for running agents in OpenShell sandboxes and lists supported agents. It offers two paths: a starter prompt for a coding agent to install it with you, and a terminal installer with express mode. The rest covers docs, community, contributing and vulnerability reporting. Install and management commands are in the quickstart on NVIDIA's site. Apache-2.0 licensed.

FAQ

What does Enter do in express install?

On supported DGX and WSL hosts it applies recommended settings and installs OpenClaw. Answering n lets you choose the agent, provider and model.

How do I check the sandbox afterwards?

Run nemoclaw <name> status; connect with nemoclaw <name> connect.

Editors’ pick

Static code analysis with rules that look like source code, plus a built-in MCP server for AI agents

CLIMedium risk16.6K

Trail of Bits skills

Trail of Bits Skills Marketplace

Editors’ pick

Trail of Bits plugin marketplace for security work: smart contracts, C/C++ and Rust review, Semgrep, CodeQL and fuzzing

PluginMedium risk7.1K

Nine skills that let an agent run Strix pentests, fix validated findings and add security scanning to CI

SkillHigh risk62.5K

A cybersecurity skill library: forensics, incident response, pentesting, cloud and malware analysis mapped to MITRE ATT&CK and NIST

SkillHigh risk32.8K