New architecture enables persistent, governed and auditable AI agent environments at enterprise scale
VAST Data has introduced a new Agent Sandbox Foundation Stack with NVIDIA to help enterprises securely deploy and scale thousands of autonomous AI agents and sub-agents in production environments.
The solution combines the VAST AI Operating System with NVIDIA NemoClaw and NVIDIA OpenShell, bringing together isolated agent execution, persistent data services and operational visibility. It addresses a key challenge in agentic AI deployments: ensuring that an agent’s memory, working data and task progress remain available even when its compute environment is terminated or replaced.
As AI agents increasingly execute complex tasks, spawn sub-agents and operate across multiple sessions, enterprises need infrastructure that separates temporary execution from persistent state. The new stack enables agents to pause, resume or move across eligible infrastructure without losing their progress or context.
“Every enterprise that puts an agent into production will eventually need to answer what it did and why it did it,” said Alon Horev, CTO and Co-Founder of VAST Data.
The architecture combines dynamically provisioned workspaces using VAST zero-copy clones, persistent storage and shared data services through VAST DataStore and VAST DataBase, and real-time operational visibility. It also maintains durable records of agent activity, including tool calls, prompts, token usage, costs and decisions, supporting governance and auditability.
The announcement extends VAST Data’s Foundation Stacks initiative, which aims to turn NVIDIA AI Blueprints into repeatable, production-ready implementations.
VAST plans to demonstrate the stack at NVIDIA GTC Berlin through a multi-agent workflow showing secure execution, persistent data access and continuity across changing compute environments. The company also plans to make automated deployment available through a public GitHub repository, enabling organizations to reproduce the architecture in their own environments.
