Hewlett Packard Enterprise announced the integration of Nvidia OpenShell, part of the Nvidia Agent Toolkit, into its HPE Private Cloud AI platform at HPE Discover in Las Vegas on June 16. The combined system is designed as an agent operating system, providing behavior monitoring, policy enforcement, sandboxing (isolating actions in a controlled environment), rogue-agent detection and rollback through HPE’s Zerto software. OpenShell is an open-source framework available on GitHub and applies controls without requiring changes to an AI agent’s code. HPE also added Nvidia Vera CPU support to its latest HPE ProLiant Compute DL394 Gen12 servers, while the platform now supports up to 256 GPUs in multi-node inference configurations. The phased rollout begins in July 2026, with full Agent Toolkit support expected by Q4 2026. HPE is positioning the platform’s sovereign, on-premises deployment model as a way to address governance requirements linked to transparency, accountability and risk management under the EU AI Act.