Alon Horev Co Founder VP Technology
Enterprises running production systems require dedicated AI agent sandboxes to operate autonomous software without risking data corruption or operational disruption. VAST Data and NVIDIA introduced the Agent Sandbox Foundation Stack on October 22, 2026, combining the VAST AI Operating System with NVIDIA NemoClaw and NVIDIA OpenShell. The unified system gives organizations isolated workspaces alongside persistent data storage for managing fleet operations.
As autonomous workloads expand, fleets frequently generate and terminate thousands of execution environments within an hour. Individual agents execute specific tasks, spawn short-lived subagents, and shut down upon completion. Consequently, enterprise infrastructure must preserve memory, execution history, and state across distributed compute resources even when individual compute nodes terminate.
Architecture for AI agent sandboxes
The joint architecture establishes isolated environments while maintaining durable records across the entire software lifecycle. Specifically, NVIDIA OpenShell provides runtime boundary enforcement and policy controls around automated actions. Meanwhile, VAST Data supplies scalable storage services, integrating data pipelines directly with AI agent sandboxes to preserve intermediate data.
“Every enterprise that puts an agent into production will eventually need to answer what it did and why it did it. An agent might exist for minutes, spawn thousands of other agents and then disappear, but its work has to persist and its actions have to stay understandable long afterward.”
Alon Horev, CTO and Co-Founder at VAST Data
Separating Persistent State from Ephemeral Compute
To support high-frequency workload cycles, the system uses zero-copy clones that rapidly create dedicated storage spaces without generating full duplicates of base datasets. VAST DataStore maintains persistent snapshots and sandbox storage, whereas VAST DataBase enables multi-agent access to shared enterprise contexts. As a result, agents can pause, transfer, or resume tasks across different computing systems without losing operational context.
“As AI agents take on more complex work, enterprises need infrastructure that can support secure execution while preserving the data, context and state agents depend on across their lifecycle.”
Kari Briski, Vice President of Generative AI Software at NVIDIA
Real-Time Observability and Audit Tracking
Autonomous activity requires continuous oversight to ensure system reliability and cybersecurity compliance. The VAST Management System delivers real-time operational tracking across infrastructure resources. Concurrently, VAST DataBase logs prompts, tool invocations, token usage, computing costs, and final decisions. This mechanism helps platform administrators supervise active AI agent sandboxes across distributed server clusters.
Demonstration at NVIDIA GTC Berlin
The stack extends earlier foundation implementations created for enterprise research assistance and video intelligence. In addition, VAST Data will demonstrate the complete software workflow at the NVIDIA GTC Berlin conference on modern computers and enterprise servers. Furthermore, the company plans to release the deployment automation files through a public GitHub repository for enterprise apps and developer teams.
