Hewlett Packard Enterprise integrates Nvidia OpenShell into private cloud AI platform
The new integration creates an 'agent operating system' designed to monitor, sandbox, and roll back autonomous AI agents in enterprise environments.
Hewlett Packard Enterprise just gave its private cloud AI platform a significant upgrade, and the centerpiece is something that sounds deceptively simple: a secure runtime for AI agents. HPE announced the integration of Nvidia OpenShell, part of the Nvidia Agent Toolkit, into its HPE Private Cloud AI platform during the HPE Discover event in Las Vegas on June 16.
What the integration actually does
The combined HPE-Nvidia stack creates what the companies are calling an “agent operating system.” That phrase gets thrown around loosely in tech, but here it refers to a specific set of capabilities: behavior monitoring, policy enforcement, action isolation through sandboxing, and rollback functionality powered by HPE’s Zerto software.
OpenShell itself is an open-source framework available on GitHub. One notable design choice: it enforces controls without requiring code changes to the AI agents themselves.
The platform also picked up rogue-agent detection capabilities. If an AI agent starts behaving outside its expected parameters, the system flags it. Combined with continuous data protection, the idea is to catch problems before they cascade.
On the hardware side, HPE introduced support for the Nvidia Vera CPU on its latest HPE ProLiant Compute DL394 Gen12 servers. The platform now supports up to 256 GPUs in multi-node inference configurations.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
Timing and the regulatory backdrop
The phased rollout begins in July 2026, with full Agent Toolkit support expected by Q4 2026.
The EU AI Act has been reshaping how companies think about deploying AI systems, particularly autonomous ones that make decisions without human oversight. HPE is positioning its Private Cloud AI platform as an answer to governance requirements around transparency, accountability, and risk management. The sovereign and on-premises deployment model means data stays within the organization’s control.
The sandboxing capability isolates agent actions so that even if an AI system behaves unexpectedly, the blast radius is contained. The rollback capability via Zerto lets administrators revert to a known-good state.