NVIDIA Open-Source AI Agent Platform

Run NemoClaw — Safer AI Agents, One Command

Secure OpenClaw with sandboxing, policy-based inference, and enterprise guardrails. NVIDIA OpenShell and Nemotron. Open source. Get started in minutes.

Get started View on GitHub
Overview

What Is NemoClaw?

NemoClaw is an open-source stack from NVIDIA that simplifies running OpenClaw always-on assistants safely — with policy-based security, sandboxing, and inference routed through NVIDIA cloud.

OpenClaw is the open-source AI assistant that lets users run LLM-powered agents locally for writing, coding, and file operations on-device. NemoClaw is NVIDIA's plugin on top: it installs the NVIDIA OpenShell runtime (part of the Agent Toolkit), creates a sandboxed environment, and routes every inference call through declarative policy. The result is a secure, controllable way to run OpenClaw with enterprise-grade guardrails.

NemoClaw uses a versioned blueprint to orchestrate sandbox creation, network and filesystem policy, and inference setup. The nemoclaw CLI handles the full stack — OpenShell gateway, sandbox, inference provider, and network policy — so you can deploy with one command and chat with your agent via TUI or CLI.

The platform integrates with the NVIDIA NeMo framework, Nemotron models, and NIM (NVIDIA Inference Microservices). Inference today is routed through NVIDIA cloud (e.g. nvidia/nemotron-3-super-120b-a12b); you get an API key from build.nvidia.com and supply it during nemoclaw onboard. The ecosystem supports dedicated hardware — from GeForce RTX PCs to DGX Station and DGX Spark — and continues to expand.

How It Works

Steps to Run NemoClaw

NVIDIA NemoClaw lets you deploy secure, always-on AI assistants with a single command. For the canonical source, see the official repository and NVIDIA NemoClaw page. Here’s how it works.

Step 1

Install with one command

Run the official install script in your terminal. NemoClaw installs the stack that adds privacy and security controls to OpenClaw.

curl -fsSL https://nvidia.com/nemoclaw.sh | bash
Step 2

Onboard your agent

Start the agent and complete onboarding. You can do this from the terminal or by asking an existing AI agent to install NemoClaw for you.

nemoclaw onboard

Or ask any coding agent: “Help me install nvidia.com/nemoclaw”.

Step 3

Security and guardrails

NemoClaw uses NVIDIA Agent Toolkit to secure OpenClaw and installs NVIDIA OpenShell — an open source runtime that enforces policy-based privacy and security guardrails. You control how agents behave and handle data.

Step 4

Run models locally or via cloud

NemoClaw evaluates your compute resources to run high-performance open models like NVIDIA Nemotron™ locally for better privacy and cost. A privacy router can also connect agents to cloud-based frontier models within your defined guardrails.

You can run always-on, self-evolving agents on dedicated hardware — including NVIDIA GeForce RTX™ PCs/laptops, RTX™ PRO workstations, and DGX Station™ or DGX Spark™. For the official source and full CLI reference: github.com/NVIDIA/NemoClaw and NemoClaw docs.

Core Capabilities

NemoClaw Key Features

What you get with NemoClaw: sandboxed OpenClaw, policy-driven inference, and protection layers that keep agents under your control.

Sandbox & Protection Layers

Network egress, filesystem access, and process syscalls are governed by declarative policy. OpenShell blocks unlisted hosts and surfaces requests for approval. Sandbox baseline is locked at creation; network and inference policies can be hot-reloaded at runtime.

Open-Source & Blueprint-Driven

Versioned Python blueprints orchestrate sandbox creation, policy, and inference. The nemoclaw CLI and plugin commands are open source on GitHub. You can inspect, customize, and extend the stack without lock-in.

One-Command Install & Onboard

Run the install script, then nemoclaw onboard to create a sandbox, configure inference (NVIDIA cloud), and apply security policies. Connect with nemoclaw <name> connect and chat via OpenClaw TUI or CLI.

NVIDIA Inference & Ecosystem

Inference is routed through the OpenShell gateway to NVIDIA cloud (e.g. nvidia/nemotron-3-super-120b-a12b). NemoClaw is part of the NVIDIA Agent Toolkit and integrates with NeMo and NIM for a full AI software stack.

Ecosystem

Enterprise Partners & Ecosystem

NemoClaw fits into a broader ecosystem of enterprise AI tools and partnerships built around NVIDIA's Agent Toolkit, NeMo, and NIM.

Organizations can combine NemoClaw with existing enterprise software, security tools, and cloud infrastructure. The project is open source and actively developed; documentation and community support are available via GitHub and NVIDIA docs.

Salesforce

Global leader in CRM software. Integration with NemoClaw enables AI-powered customer relationship automation, intelligent sales workflows, and autonomous service agent deployment.

Cisco

Network and security solutions giant. Partnership focuses on embedding NemoClaw agents into enterprise network management and cybersecurity operations.

Google (Alphabet)

Cloud computing and AI powerhouse. Collaboration with Google signals potential cloud-native NemoClaw deployment options and cross-platform AI agent interoperability.

Adobe

Leading creative and document software company. Integration targets AI-driven content generation, creative workflow automation, and intelligent document processing.

CrowdStrike

Cybersecurity enterprise leader. Partnership aims to bring NemoClaw's autonomous agent capabilities to threat detection, incident response, and security operations automation.

Built-In Enterprise Security

NemoClaw ships with enterprise-grade security, privacy, and permission control systems — directly addressing the data leakage and model misuse concerns that enterprises face when deploying AI agents.

Competitive Analysis

NemoClaw vs. OpenClaw

Understanding how NemoClaw differentiates itself from OpenClaw is essential to evaluating its position in the enterprise AI agent market.

Attribute OpenClaw NemoClaw (NVIDIA)
Developer Peter Steinberger (individual project) NVIDIA Corporation
Status Acquired by OpenAI Available — open source
Target Market General-purpose consumer AI assistant Enterprise AI agent platform
Core Strength Rapid deployment, viral adoption Security, privacy, enterprise reliability
Ecosystem Community-driven (NanoClaw variants) NVIDIA NeMo & NIM integration
Governance Transitioning to foundation management NVIDIA-backed with open-source access
GPU Acceleration Not natively optimized Native NVIDIA GPU acceleration
Industry Impact

Why NemoClaw Matters

NemoClaw brings enterprise-grade safety and control to the OpenClaw ecosystem — sandboxing, policy-based inference, and NVIDIA's Agent Toolkit in one open-source stack.

AI agents that can reason, plan, and act are a step change from chatbots. OpenClaw proved the demand for local, always-on assistants; NemoClaw adds the security and governance that organizations need. With NemoClaw you get a sandboxed environment where every network request, file access, and inference call is governed by declarative policy — and inference is routed through NVIDIA cloud so model calls stay controlled and auditable.

For teams evaluating AI agent deployment, NemoClaw offers a clear path: open-source transparency, integration with NeMo and NIM, protection layers (network, filesystem, process, inference), and a single-command install. The project is in active development; the community and documentation are growing. Get started with the official repository and setup guide.

The Claw Family

Explore the Claw AI Agent Ecosystem

The OpenClaw phenomenon triggered a "big bang" in AI agents, spawning specialized variants built by communities and tech giants alike. The ecosystem now spans from ultra-lightweight edge agents to enterprise-grade platforms.

Claw Ecosystem Overview

Full comparison of all seven Claw variants — OpenClaw, NemoClaw, NanoClaw, PicoClaw, ZeroClaw, NullClaw, and IronClaw — with detailed analysis of security, hardware reach, and ecosystem capabilities.

NemoClaw vs OpenClaw

Head-to-head comparison between NVIDIA's enterprise platform and the original viral AI assistant across 11 dimensions including security, ecosystem, hardware support, and governance.

NanoClaw — Secure Sandbox Agent

Community-built container-native variant with OS-level isolation. Runs each Skill in sandboxes to prevent RCE attacks — the best personal alternative for security-conscious users. ~400MB memory.

PicoClaw — Embedded AI Agent

Built by Sipeed in Go for RISC-V and ARM boards. Runs on $10 hardware with under 10MB memory. Features Whisper voice integration for physical AI assistants and IoT deployments.

ZeroClaw, NullClaw & IronClaw — High-Performance Agents

The performance-focused end of the Claw ecosystem. ZeroClaw (Rust) starts in under 10ms with 8MB memory. NullClaw (Zig) runs in under 1MB. IronClaw (C++) delivers industrial-grade stability for long-term operation. Ideal for edge computing and VPS deployments.

NemoClaw vs Enterprise AI Agents

How NemoClaw compares to Salesforce Agentforce, Microsoft Copilot Agents, Anthropic Claude Agents, and Zhipu AutoClaw — a competitive analysis across positioning, security strategy, and open-source status.

Frequently Asked Questions

NemoClaw FAQ

What is NemoClaw?

NemoClaw is NVIDIA's open-source stack for running OpenClaw always-on assistants safely. It installs the OpenShell runtime, creates a sandboxed environment with policy-based security, and routes inference through NVIDIA cloud. It integrates with the NeMo framework and NIM so you get enterprise-grade control and data governance.

How does NemoClaw differ from OpenClaw?

OpenClaw is the AI assistant; NemoClaw is NVIDIA's plugin that adds secure installation, the OpenShell runtime, sandboxing (Landlock, seccomp, netns), and inference routed through NVIDIA cloud with declarative policy. You run OpenClaw inside NemoClaw's controlled environment.

How do I install NemoClaw?

Run curl -fsSL https://nvidia.com/nemoclaw.sh | bash, then nemoclaw onboard to configure your sandbox and inference (you'll need an NVIDIA API key from build.nvidia.com). See How to setup and the GitHub repo for full steps and prerequisites (e.g. Linux Ubuntu 22.04+, Node.js 20+, Docker, OpenShell).

Is NemoClaw open source?

Yes. NemoClaw follows open-source principles, allowing enterprises to access the platform's source code and customize AI agent behavior, workflows, and integrations according to their specific requirements — without dependency on proprietary commercial APIs.

What tasks can NemoClaw agents automate?

NemoClaw agents can be deployed across an organization to automate routine business operations including email processing, calendar and scheduling management, data analysis, report generation, and cross-system workflow orchestration — driving measurable productivity improvements at scale.

References & Sources

Further Reading