Nvidia wraps its NemoClaw around OpenClaw for the sake of security

'OpenClaw is the operating system for personal AI,' insists Nvidia CEO

gtc In Pixar's Toy Story, a trio little green aliens explain, "The claw chooses who will go and who will stay." The claw in that instance was a mechanical claw in a vending machine. 

In Nvidia's tech narrative, the claw chooses the tools it will use to carry out data processing directives and network traversal. Nvidia is referring to the trending shorthand for software agents, which are AI models given access to software tools and services. At its GTC conference in San Jose, CA, on Monday, the company unveiled NemoClaw, a set of software tools meant to help corporate customers use the OpenClaw platform.

"Claws are autonomous agents that can plan, act, execute tasks on their own, and they've gone from just thinking and executing on tasks to achieving entire missions," said Kari Briski, VP of generative AI software for enterprise at Nvidia, during a media briefing on Sunday ahead of the conference. "We used to prompt with what, how, or why, but for claws now we prompt with build, create or make."

The claw terminology dates back to the end of January with the debut OpenClaw, an open platform for connecting software agents to various applications with few constraints. (It had previously been known as Clawd, then Moltbot.) OpenClaw achieved social media fame for showing people how easy it could be to enable automation, albeit with uncertain results and multiple security nightmares. Shortly thereafter, OpenAI "acquihired" OpenClaw founder Peter Steinberger and said the project would end up being managed by a foundation.

Now Nvidia is looking to arm itself with OpenClaw, which Nvidia CEO Jensen Huang describes as "the operating system for personal AI."

Briski was more blunt. "Claws are the new application layer for AI, and they're driving orders of magnitude more demand for compute."

Nvidia, the provider of picks-and-shovels for the AI boom, wants to capitalize on that demand. But the excitement of letting AI agents muck about in corporate databases, personal correspondence, and the enterprise network tends to be a bit much for lawsuit-averse companies. As Briski observed, "Claws are exciting but they're risky too, because they could access sensitive data, misuse connected tools, or escalate privileges autonomously."

Just as NanoClaw has made a home in Docker's Sandbox, Nvidia aims to tamp down the excitement to an enterprise-compatible level.

Nvidia's NemoClaw stack lets users of OpenClaw install Nvidia's Nemotron models and OpenShell runtime using a single command via the Nvidia Agent Toolkit, a collection of models, runtimes, and blueprints for safer, long-running agents.

Briski describes OpenShell as an open-source safety and security runtime for agents. It sandboxes OpenClaw agents to limit their access to sensitive data and reduce the opportunity for unwanted behavior.

"OpenShell provides the missing infrastructure layer beneath clause to give them the access they need to be productive while enforcing policy based security, network and privacy guardrails," she explained.

NemoClaw for OpenClaw supports local computing for AI agents on PCs with Nvidia GeForce RTX, workstations equipped with Nvidia RTX Pro, Nvidia DGX Station, and DGX Spark supercomputers. ®