NVIDIA open-sources OpenShell reference architecture to secure and govern AI agent workloads
NVIDIA announced OpenShell, an open-source runtime and reference architecture that enforces least-privilege access and governance for long-running autonomous AI agents via components called Gateway, Supervisor, Sandbox, and a policy prover. NVIDIA positions OpenShell to run on its Vera CPUs, which it says are architected (Olympus cores, second-gen Scalable Coherency Fabric, LPDDR5X, PCIe Gen 6, confidential computing) to meet performance, bandwidth, I/O, and security needs of large-scale agent workloads.