The system helps companies run AI agents, monitor their actions, stop unsafe activity, and protect systems across data centres and edge devices.
NVIDIA has announced the NVIDIA OpenShell Reference Architecture, an infrastructure designed to help enterprises deploy, monitor, and secure AI agents running on enterprise infrastructure and edge systems. It combines hardware and software to contain agent activity, monitor behaviour, and enforce security policies during execution.
The architecture extends NVIDIA OpenShell, an open-source runtime for AI agents, by combining NVIDIA Vera Rubin infrastructure, NVIDIA Vera central processing units (CPUs), and NVIDIA BlueField data processing units (DPUs). Vera Rubin runs AI models and workloads, including open and closed models built with NVIDIA Nemotron, while Vera CPUs run agent harnesses. BlueField DPUs run NVIDIA Sentry, a security framework built on NVIDIA DOCA software, to monitor agent behaviour and enforce policies in silicon.
If an agent operates outside its permitted software boundary, Sentry can quarantine and stop it within milliseconds, according to NVIDIA. The company claims that Sentry running on BlueField-4 can increase agent capacity by up to 20%. NVIDIA also claims that Vera Rubin NVL72 infrastructure can deliver up to 30 times higher throughput per megawatt for agentic workloads, while Vera CPUs offer twice the performance of x86 processors with greater energy efficiency.
The architecture also supports AI agents deployed at the edge in sectors such as retail, healthcare, manufacturing, robotics, software development, and telecommunications. Supported systems include NVIDIA RTX Spark, NVIDIA DGX Spark, and NVIDIA DGX Station.
NVIDIA is working with infrastructure providers, model developers, software companies, and cybersecurity firms to support deployments. The architecture supports Linux and Kubernetes environments from Canonical, SUSE, and Red Hat, as well as Microsoft Windows through Microsoft Execution Containers. It can also run on cloud operating systems from Amazon Web Services, Google Cloud, and Microsoft Azure.
“AI capability and safety must advance together,” said Jensen Huang, NVIDIA’s founder and CEO. “As enterprises prepare to deploy billions of AI agents, safety and control must be built into the computing architecture. NVIDIA created OpenShell to give agents the freedom to do useful work within clearly defined boundaries — contained, monitored and under control. With the OpenShell Reference Architecture, agents can think on Rubin, act on Vera and be governed on BlueField.”
Click here for the original announcement.
