THE CRUNCH
NVIDIA has donated OpenShell, an open source runtime for securing and governing autonomous AI agents, to the Cloud Native Computing Foundation, where it enters the CNCF Sandbox programme. Announced at KubeCon, the move hands control of agent security infrastructure to the community that builds and inspects it. OpenShell applies deterministic controls to what agents can do, letting teams enforce policies as agents use tools, access inference services and run workloads, while its extensible architecture connects to existing compute, credential storage and monitoring systems. NVIDIA says hundreds of organisations already use it, with tens of thousands of contributions in under a year since launch.
The donation sits alongside a contribution to Agent Substrate, a Google-originated CNCF Sandbox framework for running long-running agent workloads on Kubernetes. It preserves agent environments between tasks and releases compute when agents are idle, so teams can scale without dedicated hardware for every session. NVIDIA is ensuring it works with its Dynamo inference serving and KAI Scheduler tools, and is collaborating with Google on an OpenShell and Agent Substrate integration spanning clouds and on-premises systems.
NVIDIA also announced three new open source components of its DSX AI Factory Platform. NVIDIA Kubernetes Engine and NVIDIA Kubernetes Recipes, both in public preview, automate cluster lifecycle management on GPU infrastructure with signed, version-pinned images. NVIDIA Snapshot restores checkpointed inference replicas in seconds without reloading models or warming up workloads, easing the delay of bringing extra capacity online.
Early users report striking speed-ups from Snapshot. Crusoe, an upstream contributor, tested it on GLM-5.3 running on Crusoe Cloud with NVIDIA HGX B300 systems and cut the workload's time to ready from roughly eight minutes to 31 seconds. Saturn Cloud benchmarked scale-from-zero inference and reduced vLLM startup times from between two and 8.5 minutes to between six and 34 seconds.
Mirantis, Netris, OpenNebula Systems, Rafay, Spectro Cloud and vCluster, along with Firmus and SK Telecom, have aligned with NVIDIA on integration plans to bring the Kubernetes Engine and Recipes capabilities into their offerings.