At KubeCon + CloudNativeCon, NVIDIA announced that it is contributing OpenShell to the Cloud Native Computing Foundation (CNCF). The foundation is a nonprofit within the Linux Foundation. OpenShell is an open source runtime for securing and governing autonomous agents, and it enters the CNCF as a Sandbox project.
What OpenShell does
The runtime applies deterministic controls to what an agent does, so a team can enforce policies while the agent calls tools, reaches inference services or runs workloads. The architecture is extensible. Teams can plug in their own compute, inference providers, credential storage and monitoring, which lets them apply the same governance across different environments and connect it to existing enterprise systems.
NVIDIA says hundreds of organizations have used OpenShell and that the project has drawn tens of thousands of contributions in under a year since launch.
Agent Substrate and Google
NVIDIA is also contributing to Agent Substrate, a Google-originated framework, already in the CNCF Sandbox, for running long-lived agent workloads at scale on Kubernetes. NVIDIA is making it work with its Dynamo inference software and the KAI Scheduler, which allocates shared cluster resources. Agent Substrate keeps an agent's environment between tasks and releases compute while the agent is idle, so a team does not need dedicated compute for every session. NVIDIA and Google are also working on an integration of OpenShell and Agent Substrate for governing agents across clouds and on-premises systems. That integration is in progress and has no release date in the source.
New Kubernetes tools for AI factories
Three open source components extend NVIDIA's DSX AI Factory Platform:
- NVIDIA Kubernetes Engine (NVKE) and Kubernetes Recipes (NVKR) are in public preview. They automate cluster lifecycle management and supply signed, version-pinned images for providers that sell managed Kubernetes on GPU infrastructure. Mirantis, Netris, OpenNebula Systems, Rafay, Spectro Cloud and vCluster, along with Firmus and SK Telecom, have agreed integration plans.
- NVIDIA Snapshot restores checkpointed inference replicas in seconds, skipping model reload and warmup. Crusoe, testing GLM-5.3 on its cloud with NVIDIA HGX B300 systems, cut time to readiness from roughly eight minutes to 31 seconds. Saturn Cloud cut vLLM startup from between two and 8.5 minutes to between six and 34 seconds in scale-from-zero tests.
These are vendor-reported results from design partners on their own workloads.
What is and is not settled
OpenShell's governance model is now shared. Organizations building it into their products will steer it, which the source presents as a way to strengthen its safeguards. The source gives no pricing and no timeline for production release of the NVKE and NVKR previews. Those two tools serve providers running GPU infrastructure, not companies that only consume AI services.