A secure orchestration plugin for running always-on AI assistants in policy-enforced sandboxes using NVIDIA OpenShell.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install nvidia-nemoclaw
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install nvidia-nemoclaw using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is NVIDIA NemoClaw?
NVIDIA NemoClaw is a professional-grade TypeScript CLI plugin designed to simplify the installation and management of secure, always-on AI assistants. As a specialized entry in the Openclaw Skills collection, it leverages the NVIDIA OpenShell runtime to create isolated environments where agents can operate safely. It handles everything from sandbox creation using Landlock and seccomp to routing inference through NVIDIA cloud Nemotron models, ensuring that all agent activities are governed by strict, declarative security policies.
NVIDIA NemoClaw Use Cases
- Deploying enterprise-grade AI assistants that require restricted network access to prevent unauthorized data egress.
- Creating secure testing environments for experimental Openclaw Skills using policy-enforced sandboxes.
- Orchestrating multiple AI agents across local hardware or remote GPU instances via Brev integration.
- Building custom automated workflows that require programmatic control over AI agent lifecycle and security perimeters.
How NVIDIA NemoClaw Works
- The onboard command initiates an interactive wizard to configure the sandbox name, API keys, and initial security policies.
- The system resolves the required blueprints and verifies resource digests to ensure environment integrity.
- NemoClaw provisions a secure sandbox via OpenShell, isolating filesystem access, syscalls, and network namespaces.
- All inference calls made by the internal OpenClaw agent are routed through a managed gateway to NVIDIA cloud services.
- Developers manage the agent through the host CLI or a programmatic API, with the ability to monitor logs and adjust policies in real-time.
NVIDIA NemoClaw Setup
To get started with this addition to your Openclaw Skills, ensure you have Linux Ubuntu 22.04+, Node.js 20+, Docker, and NVIDIA OpenShell installed. Use the one-line installer for rapid deployment:
curl -fsSL https://nvidia.com/nemoclaw.sh | bash
For manual installation from source:
git clone https://github.com/NVIDIA/NemoClaw.git
cd NemoClaw
npm install
npm run build
npm link
Finally, export your NVIDIA API key:
export NVIDIA_API_KEY="nvapi-xxxxxxxxxxxx"
NVIDIA NemoClaw Data Schema & Taxonomy
NemoClaw manages its environment and agent metadata through a structured schema to ensure consistency across the Openclaw Skills ecosystem:
| Component | Description |
|---|---|
| Sandbox Metadata | Contains the unique ID, health status, and active blueprint version. |
| Policy Configuration | A declarative JSON structure defining allowed egress hosts and filesystem paths. |
| Inference Profile | Metadata regarding the specific Nemotron model and cloud routing settings. |
| Blueprint Artifacts | Versioned Python scripts that define the resource plan for the sandbox environment. |
By default, sandbox data is stored in /var/nemoclaw/sandboxes unless the NEMOCLAW_SANDBOX_DIR environment variable is specified.
NVIDIA NemoClaw Advanced Features
- Programmatic TypeScript API for building complex, automated AI agent integrations.
- Hot-reloadable network policies that allow for real-time adjustments to egress rules without agent downtime.
- Support for remote deployment to specialized GPU instances via Brev.dev integration.
- Integrated TUI via OpenShell for manual approval of blocked network requests and deep inspection of agent behavior.
- Multi-layer security encompassing network isolation, filesystem locking, and syscall filtering.
SKILL.md
Loading
Related Openclaw Skills
Agent Architecture Evaluator
A comprehensive diagnostic tool for reviewing, testing, and optimizing the structural integrity of complex agentic architectures and multi-agent systems.
pidrive
A CLI-based storage solution that mounts AWS S3 as a local filesystem via WebDAV, enabling AI agents to manage files using standard Unix commands.
SkillBoss Multi-Model AI Hub
Access 50+ AI models for chat, media generation, and developer tools via a single API key through Openclaw Skills.
SkillBoss: Universal AI API Hub
A unified API gateway providing access to over 50 leading AI models for text, image, video, audio, and specialized automation tasks through a single interface.
obra/superpowers Agentic Skills Framework
A composable skills framework that provides coding agents with structured development workflows including TDD, systematic debugging, and subagent-driven execution.
ProductClank Campaigns
Turn your community into a growth engine with AI-powered social media amplification and authentic conversation discovery.
Featured*
