# NVIDIA donates its OpenShell agent runtime to the CNCF and open-sources Snapshot for faster inference restarts

> OnPrem World · https://onpremworld.com/en/news/2026-11-nvidia-openshell-cncf-snapshot · CC BY 4.0

Published: 2026-11-10 · Other

At KubeCon + CloudNativeCon North America on November 10, NVIDIA said it is contributing OpenShell, its open source runtime for securing and governing autonomous AI agents, to the Cloud Native Computing Foundation as a Sandbox project. OpenShell applies deterministic policy controls as agents use tools, call inference services and run workloads, and NVIDIA and Google are integrating it with Agent Substrate, a Google-led framework for long-running agent workloads on Kubernetes. NVIDIA also released Snapshot, an open source component that restores checkpointed inference replicas in seconds without reloading or warming up the model, and put NVIDIA Kubernetes Engine (NVKE) and Kubernetes Recipes (NVKR) into public preview.

## Why it matters locally

NVIDIA says the OpenShell and Agent Substrate integration is meant to govern agents across clouds and on-premises systems, so teams running agents and inference on their own Kubernetes GPU clusters get vendor-neutral, open source building blocks for both.

## For the market

Handing OpenShell to a neutral foundation makes it easier for other vendors to adopt, and agent governance could become a standard layer the way Kubernetes did for containers. It also keeps NVIDIA's software, such as Dynamo, close to the default path for anyone building agent platforms.

## In plain words

AI agents are programs that can act on their own, like sending files or running tools. OpenShell works like a building's security desk: it checks every action against the company's rules before letting the agent through, and it is now free and shared by the whole industry.

## For practitioners

NVKE and NVKR are public previews, so test them on a staging cluster before production. Snapshot targets slow cold starts: in NVIDIA's examples, Saturn Cloud cut vLLM startup from between two and 8.5 minutes to between six and 34 seconds, and Crusoe cut a GLM-5.3 workload's time to ready from roughly eight minutes to 31 seconds on HGX B300 systems. Measure on your own models and GPUs, since those were partner results on data center hardware.

## Source

[NVIDIA blog: NVIDIA Expands Open Source Commitment With OpenShell Donation and New AI Infrastructure](https://blogs.nvidia.com/blog/open-source-openshell-infrastructure/)

[All news](https://onpremworld.com/en/news.md)

Summaries are written by OnPrem World from the official source. We link to the original and never republish it.
