NVIDIA has introduced a new AI stack, NemoClaw, for the OpenClaw agent platform, designed to enhance the security, privacy, and control of autonomous AI agents.
The announcement was made yesterday at the company’s GPU Technology Conference, where NVIDIA said the system is designed to make self-evolving AI agents more deployable and scalable, particularly in enterprise environments.
The pitch centres on making these agents usable beyond experimentation. OpenClaw’s agents, often called “claws,” have gained attention because they can handle a wide range of tasks inside virtual environments, using different tools and models. That flexibility also creates risk. Without clear guardrails, the same agents can access sensitive data or systems in ways that make organisations uneasy.
NemoClaw introduces a layer designed to manage that risk. It installs an environment called OpenShell that runs agents in a sandbox. That setup isolates what the agent can see and do, while allowing developers to define rules around data access, network activity, and behaviour.
NVIDIA describes it as the missing infrastructure piece with autonomous agents. The company says the system can be deployed with a single command, combining its Agent Toolkit with support for open models like NVIDIA Nemotron.
“OpenClaw opened the next frontier of AI to everyone and became the fastest-growing open source project in history,” said Jensen Huang, NVIDIA’s CEO. “OpenClaw is the operating system for personal AI.”
AI agents are gradually moving from tools that respond to prompts to systems that act continuously, running in the background, making decisions, and interacting with software on behalf of users.
To support that, NemoClaw splits how models are used. Some can run locally on dedicated machines, while others can be accessed through the cloud using what NVIDIA calls a “privacy router.” The idea is to give agents flexibility without exposing sensitive data unnecessarily.
This hybrid setup reflects where the industry is heading. Developers want the power of large cloud models, but enterprises still prefer keeping certain workloads local. NemoClaw attempts to sit between those needs.
Damilare Odedina
