NVIDIA has introduced NemoClaw, a new open-source software stack designed to securely deploy autonomous AI agents, during the keynote at the NVIDIA GTC 2026 conference in San Jose, California.

Announced on March 16 by Jensen Huang, the company’s founder and CEO, NemoClaw expands the rapidly growing OpenClaw agent framework by adding enterprise-grade privacy, security, and policy controls.

NVIDIA described the platform as a reference implementation that helps organizations deploy always-on AI agents safely, addressing growing concerns about security and data governance in agent-based systems.

Making AI Agents Enterprise-Ready

NemoClaw builds on the OpenClaw framework—an emerging platform for creating autonomous AI agents sometimes referred to as “claws.”

While OpenClaw has gained significant traction among developers, enterprises have been cautious due to concerns about: data privacy, uncontrolled access to systems, and regulatory compliance.

NemoClaw aims to solve those issues by introducing policy-driven guardrails and sandboxed environments that control how AI agents interact with networks, files, and external services.

Key Security and Infrastructure Features

The NemoClaw stack includes several components designed to improve the safety and manageability of agentic AI deployments.

Key features include:

• Policy-based guardrails to restrict data access and system actions

• OpenShell runtime, which sandboxes agent execution

• Privacy routing, allowing organizations to choose between local models or cloud-based inference

• Hybrid deployment, enabling agents to run locally, on-premises, or in the cloud.

The platform can operate across a wide range of hardware environments, including NVIDIA’s RTX GPUs, DGX Station, and DGX Spark AI systems.

Simple Installation for Developers

NVIDIA highlighted the platform’s ease of use, allowing developers to deploy the stack using a single command-line installation followed by a guided onboarding process.

Once installed, developers can configure and run AI agents capable of performing tasks such as: coding assistance, workflow automation, research and analysis, and integration with developer tools like Ollama.

The system also supports local inference using open models such as the Nemotron series, reducing reliance on external cloud services.

Part of NVIDIA’s Expanding AI Ecosystem

NemoClaw forms part of NVIDIA’s broader Agent Toolkit, which combines: the NemoClaw security layer, the OpenShell runtime environment, and NVIDIA’s Nemotron family of AI models.

The company said the toolkit is intended to help organizations safely deploy “agentic AI” workflows, where autonomous systems handle complex tasks across software environments.

Early Preview and Developer Access

NemoClaw is currently available in an early preview stage, with source code and documentation released publicly for developers.

Participants attending GTC 2026 can also experiment with the platform through the “Build-a-Claw” workshop, where developers create customized agents running on DGX workstations and RTX-powered laptops.

NVIDIA’s launch of NemoClaw reflects the growing industry shift toward autonomous AI agents capable of performing complex tasks independently. By adding security guardrails and enterprise controls to the OpenClaw ecosystem, NVIDIA is positioning itself not only as a leader in AI hardware but also as a key provider of infrastructure for the next generation of agent-driven software systems.

‍