Nvidia bets on OpenClaw, but adds a security layer - how NemoClaw works

At its GTC keynote, Nvidia CEO Jensen Huang positioned OpenClaw as a pivotal development and introduced NemoClaw to tackle the platform's security shortcomings. OpenClaw itself does not run a single model; it combines the differing strengths of Anthropic's Claude and OpenAI's ChatGPT while running locally and taking autonomous actions — a capability that has impressed users but raised significant privacy and security concerns.

NemoClaw combines Nvidia's Agent Toolkit with a new runtime called OpenShell to add an infrastructure layer beneath OpenClaw. OpenShell sandboxes models, enforces policy-based guardrails, bolsters data privacy protections and improves scalability, and Nvidia said it built the runtime with security firms such as CrowdStrike, Cisco and Microsoft Security to ensure compatibility with existing cyber tools.

The company says NemoClaw can be installed with a single command, runs on any platform and can use any coding agent, including Nvidia's Nemotron model family on local systems.

nvidia, openclaw, nemoclaw, openshell, agent toolkit, nemotron, claude, chatgpt, sandboxing, data privacy