A planned enterprise integration
On September 28, 2026, HPE said it is integrating NVIDIA OpenShell into HPE Private Cloud AI, with availability planned for the fourth quarter of 2026. HPE describes the work as connecting AI agent execution with enterprise identity, permission approvals, monitoring and audit records. The timetable applies to HPE’s planned integration.
Source: HPE: Planned governed agentic AI integration for enterprise production
Runtime controls and infrastructure
HPE says OpenShell uses isolated sandboxes to control file access, execution and network activity outside the agent process. Operators retain authority to approve requests for changed permissions. Its description of the general availability release also includes multi-tenancy and Kubernetes-based high availability.
NVIDIA’s separate September 28 announcement describes OpenShell as broadly available. The wider Open Agent Safety Platform also includes Sentry, a reference system design for monitoring and policy enforcement on BlueField-4 DPUs. These are vendor descriptions of the architecture, rather than independent evidence that a particular deployment will meet its security requirements.
Source: HPE: Planned governed agentic AI integration for enterprise production, NVIDIA: Open Agent Safety Platform announcement
LANGOM perspective
For procurement teams, the immediate task is to make availability and support responsibilities explicit. A proposal should distinguish software that can be evaluated now from features dependent on a future integrated release. Ask for a supported configuration, a delivery milestone and acceptance conditions for the exact environment being purchased.
Start the evaluation with one bounded business workflow. Document which files, services and model endpoints the agent may use, who can approve an exception and how permissions are withdrawn. Request demonstrations of a denied action, an approved change and the resulting audit trail using representative test data.
Infrastructure sizing should follow that workflow: consider concurrent agents, workload duration, logging retention and recovery requirements before selecting compute capacity. Assign ownership for policy updates and incident handling across the application, platform and infrastructure suppliers. These are LANGOM’s evaluation considerations; the announcements do not establish pricing, delivery dates for a customer configuration or availability through LANGOM.
Sources
Related LANGOM services
Planning an equipment project?
Share the application, required configuration and delivery destination with LANGOM.
Discuss your requirements