Nvidia said a new open source tool can bring AI models back online in seconds instead of minutes. The company shared test results on November 10 as part of a set of announcements at KubeCon + CloudNativeCon.
The tool, called NVIDIA Snapshot, is part of a wider push by Nvidia into open source software for Kubernetes. Kubernetes is an open source system that manages how applications run across large computing setups, including AI data centers.
Nvidia also said it is donating its agent security software, OpenShell, to the Cloud Native Computing Foundation. The foundation is a nonprofit that is part of the Linux Foundation and hosts KubeCon.
Handing Agent Security to the Community
OpenShell is an open source runtime that controls what autonomous AI agents can do. It applies fixed rules to agent activity, so teams can enforce policies as agents use tools, reach inference services and run workloads.
The software lets teams connect their own computing power, inference providers, credential storage and monitoring tools. This allows companies to apply the same rules across different environments and link OpenShell with systems they already use.
Nvidia said hundreds of organizations have used OpenShell. The project has received tens of thousands of contributions in less than a year since launch.
As a CNCF Sandbox project, OpenShell will be governed by the organizations building it into their products. Jim Zemlin, CEO of the Linux Foundation, said the donation gives developers “a shared foundation to build on and an opportunity to shape how agent security evolves.”
Nvidia is also contributing to Agent Substrate, a Google-led framework that has been accepted into the CNCF Sandbox. The project helps companies run long-running AI agent workloads at scale on Kubernetes.
Agent Substrate keeps agent environments saved between tasks while freeing up computing power when agents sit idle. This means teams do not need to keep dedicated computing running for every session.
Nvidia said it is making sure Agent Substrate works with NVIDIA Dynamo for model inference and KAI Scheduler for sharing cluster resources. Nvidia and Google are also working on linking OpenShell and Agent Substrate so companies can secure agents across clouds and on-site systems.
New Tools for AI Factories
Nvidia announced three new parts of its DSX AI Factory Platform. They are NVIDIA Kubernetes Engine, NVIDIA Kubernetes Recipes and NVIDIA Snapshot.
The Kubernetes Engine and Kubernetes Recipes are now available in public preview. They give AI factories open source building blocks to offer managed Kubernetes services on GPU hardware, with automated cluster management and signed, version-pinned images.
Several companies have agreed with Nvidia on plans to add these tools to their own products. They include Mirantis, Netris, OpenNebula Systems, Rafay, Spectro Cloud, vCluster, Firmus and SK Telecom.
Snapshot targets a common delay in AI systems. Loading models and warming up workloads can slow down scaling and recovery, and Snapshot restores saved copies of running models in seconds without a reload or warmup.
Design partners Crusoe and Saturn Cloud tested Snapshot on their own workloads using NVIDIA HGX B300 systems. Crusoe, which also contributes code to Snapshot, tested the GLM-5.3 model on Crusoe Cloud.
Crusoe said the time for that workload to become ready fell from about eight minutes to 31 seconds. Saturn Cloud, a token factory platform, said its vLLM startup times dropped from between two and 8.5 minutes to between six and 34 seconds.
