NVIDIA today announced NemoClaw for the OpenClaw agent platform; this article references the project as nemoclaw. the stack installs NVIDIA Nemotron models and the newly announced OpenShell runtime in a single command, adding privacy and security controls designed to make autonomous AI agents—called claws—more trustworthy, scalable and accessible.
What Nemoclaw Is and How It Installs Open Models
The NemoClaw stack is built to run on the OpenClaw agent framework and uses NVIDIA Agent Toolkit software to streamline setup. It installs OpenShell to provide open models and an isolated sandbox that adds data privacy and security to autonomous agents. The stack can tap open models, including NVIDIA Nemotron models, and supports running those models locally on a user’s dedicated system or routing to cloud-based frontier models with privacy controls in place.
Security, Privacy and Sandbox Controls
NVIDIA framed NemoClaw as adding the missing infrastructure layer beneath claws: providing access for agents to be productive while enforcing policy-based security, network and privacy guardrails. Jensen Huang, founder and CEO of NVIDIA, described OpenClaw as a new operating system for personal AI and positioned NemoClaw as a tool to make self-evolving agents more trustworthy and scalable. Peter Steinberger, creator of OpenClaw, said the collaboration aims to build the claws and guardrails that let anyone create secure AI assistants.
Nvidia described nemoclaw as an early-stage alpha release and warned users to expect rough edges while work continues toward production-ready sandbox orchestration. The stack is designed to be hardware agnostic and can use any coding agent, combining local compute with cloud models a privacy router to allow agents to develop new skills within defined guardrails.
Platforms Supported and GTC Build-a-Claw Event
NemoClaw for OpenClaw can run on a range of dedicated platforms, including GeForce RTX PCs and laptops, RTX PRO-powered workstations, and DGX Station and DGX Spark AI supercomputers, providing local compute for always-on agents. NVIDIA invited GTC attendees to its build-a-claw event in GTC Park March 16-19, with sessions scheduled 1: 00 pm ET – 5: 00 pm ET on Monday and 8: 00 am ET – 5: 00 pm ET Tuesday through Thursday, where participants can customize and deploy proactive, always-on AI assistants using the NemoClaw stack.
The announcement positions NemoClaw as an enterprise-oriented layer on top of the popular OpenClaw framework, bundling model access, runtime and sandboxing into a single installation flow intended to accelerate adoption while adding safeguards for privacy and security.
