NemoClaw lets NVIDIA join the AI Agent race

NVIDIA announced NemoClaw at its GTC conference this week, and the pitch is straightforward: take OpenClaw, the autonomous AI agent framework that became the fastest-growing Open Source project in history, and wrap it in a security and privacy layer that enterprises can actually use. The result is, essentially, an enterprise-grade distribution of OpenClaw, powered by NVIDIA's very own Agent Toolkit.

NemoClaw runs OpenClaw inside an isolated OpenShell sandbox. Inference requests never leave the sandbox directly (OpenShell intercepts every call and routes it to the configured provider). On the network side, only endpoints listed in the policy are allowed; when the agent tries to reach an unlisted host, OpenShell blocks the request and surfaces it for operator approval. More details are available here.

On a broader context, we already reported on NVIDIA being the top Open Source contributor to Hugging Face last year, a figure Jensen Huang himself highlighted at GTC as evidence of a deliberate OS strategy. NemoClaw fits that pattern: Nvidia developed it in collaboration with OpenClaw founder Peter Steinberger, who maintains the project even after joining OpenAI. Building on an existing ecosystem rather than competing with it is a sensible move, probably a faster one too.

Jensen Huang said that every company will soon need an AI agent strategy, and if those agents need to run on secure compute, NVIDIA wants to be the infrastructure that makes that happen. Whether 'claws' actually reach the scale of the protocols Huang was invoking remains open for now, but NemoClaw is a credible attempt to clear the enterprise adoption hurdle that raw OpenClaw couldn't, and that alone makes it worth watching.