Nvidia Debuts Agile ONE SN Humanoid, OpenShell at GTC Berlin

At GTC Berlin, Nvidia introduced Agile ONE SN, a reference humanoid built on Isaac GR00T with a 31-degree-of-freedom body, pitched as a drop-in starting point to cut researcher setup time from weeks to days. The company said it will be available next year.

"This is a great starting point," Jensen Huang told the Tempodrom audience, framing the reference design as modular so researchers can take the full workflow or pull in selected components. The platform leans on Nvidia's existing physical AI stack: Isaac Capture for data collection, GR00T models for robot intelligence, Isaac Lab for simulation-based training, and Isaac OS for runtime. Partners already building on the same platform include Agility Robotics, Figure, KUKA and NEURA Robotics, a lineup that fills much of our recent robotics coverage.

The second thread was agent security. Nvidia announced OpenShell, an open-source, policy-controlled runtime with kernel-level sandboxing designed to isolate what an AI agent is allowed to touch. "The most important thing is to make sure the agent is secure," Huang said. "Every single time an agent is launched, the privileges are different."

Alongside OpenShell came the Agent Knowledge Foundation, a set of models and libraries for connecting agents to enterprise data, including new Nvidia Kumo models for making predictions from structured data. "Ultimately, what turns these super agents, these agents into super agents, is that you give them access to your company's knowledge," Huang said.

The keynote closed with a pitch to European buyers: start from an open model like Nemotron, post-train on proprietary data, and run what Huang called "sovereign intelligence."