How do OpenShell, Sentry, BlueField-4, DOCA, and Vera work together?

Each component serves a distinct role: OpenShell governs how agents execute, what they can access and change, and where inference runs. Sentry with BlueField-4 delivers tenant isolation outside the agent's execution environment and provides in-silicon threat detection, identity governance, and policy enforcement using DOCA. NVIDIA Vera CPUs provide high-performance, power-efficient compute for orchestration, sandboxed code execution, and data processing.

Does OpenShell require BlueField-4?

No. OpenShell can run on supported local, on-premises, cloud, and Kubernetes infrastructure without BlueField-4. On systems with BlueField-4, Sentry adds hardware-isolated monitoring and enforcement that remain operational if the host or workload is compromised.

NVIDIA Open Agent Safety Platform

Open, customizable tools that enforce boundaries around AI agents.

Read the Announcement

Get Started

Overview

Engineering Full-Stack Safety

As agents become a digital workforce, developers need tools to verify the safety of their products. Without the right engineering solutions, AI agents can take actions with unintended consequences. NVIDIA Open Agent Safety Platform is an open reference design built with partners that continuously monitors and governs agent behavior, ensuring that AI agents follow the rules.

It features NVIDIA OpenShell™, an open source runtime that provides governance tools to enforce what an agent can see, do, and interact with. The controls remain in force when agents behave unexpectedly, limiting how far mistakes spread. To extend protection into the infrastructure, NVIDIA Sentry runs on NVIDIA BlueField™ from a hardware-isolated, host-independent security domain — delivering continuous out-of-band agent monitoring, trusted telemetry, and in-silicon policy enforcement, with millisecond-scale quarantine.

All of this is optimized to run on NVIDIA Vera CPU and NVIDIA BlueField DPU-based systems, and is also compatible with other hardware systems.

Build and Run Safe AI Agents

Introducing NVIDIA OpenShell Reference Architecture, a full-stack industrial-scale infrastructure enabling safe agents across AI factories, enterprise infrastructure, and the edge.

Read the Announcement

Operate Secure AI Agents at Scale

Together, NVIDIA OpenShell secure runtime and NVIDIA Vera CPUs govern AI agents more efficiently, at industrial scale.

Read the Technical Blog

Benefits

A Layered Foundation for Agent Safety

Combine runtime governance, continuous threat detection, and hardware-isolated policy enforcement to keep AI agents contained, observable, and auditable.

Defense in Depth

Multi-layered software and hardware work together to establish a secure runtime boundary around every agent. Policies are formally verified, so agent behavior is kept in check.

Out-of-Band Agent Governance

Real-time, host-independent monitoring and enforcement maintain an independent security boundary, even if the host is compromised.

Security at AI Agent Speed

In-silicon threat detection, identity governance, and policy enforcement work together to evaluate AI agents in real time—enforcing policies directly in the infrastructure data path and responding to detected deviations.

Always-On Threat Detection

Gain continuous visibility into agent activity and behavior—combining out-of-band runtime signals with agent interaction telemetry to detect threats as they emerge, and contain them in milliseconds.

NVIDIA OpenShell Reference Architecture for Safe AI Agents

A full-stack, open architecture powered by the NVIDIA OpenShell secure runtime, Vera CPUs, Vera Rubin systems, and NVIDIA Sentry on BlueField-4.

Learn More

Technology

Explore NVIDIA Open Agent Safety Platform

The software and hardware that governs, protects, and powers AI agents at enterprise scale.

Software

NVIDIA OpenShell

Separate how your agents execute from how they reach data, tools, and outside systems.
Govern agent behavior with sandboxed execution and policy enforcement to safely scale agentic systems for your enterprise.
Enforce deterministic governance with a zero-trust architecture that grants permissions based on intent and uses out-of-process enforcement that can't be bypassed.

Learn More About OpenShell

Software

NVIDIA Sentry

Combine out-of-band runtime signals with agent interaction telemetry to detect threats without relying on host- or agent-reported logs.
Verify agent identity and delegated authority to govern access to data, models, tools, and services—enabling zero-trust governance without relying on the agent to enforce its own permissions.
Enables developers, cybersecurity teams, and security providers to integrate their own signals, policies, and detection logic across agent monitoring, governance, and enforcement.

Learn More About NVIDIA Sentry

Hardware

NVIDIA BlueField

Provide a hardware-isolated, host-independent security domain for NVIDIA Sentry to monitor and enforce policy outside the agent and host software it protects.
Accelerate security policy enforcement directly in the hardware data path, enabling millisecond-scale agent containment and quarantine.
Scale continuous agent security across AI infrastructure with BlueField’s dedicated infrastructure processing and hardware acceleration, preserving host compute resources and increasing agent capacity.

Learn More About BlueField

Hardware

NVIDIA Vera CPU

Vera is purpose-built for agentic reasoning, tool execution, and task planning—so the agent control plane lives on Vera.
Vera achieves up to 80% faster sandbox performance than traditional CPU infrastructure, making “sandbox everything” the default instead of a tradeoff.

Learn More About Vera CPU

Ecosystem

Leading Adopters Across All Industries

Industry leaders from across the AI ecosystem are joining NVIDIA to strengthen AI safety for every industry across the full stack of infrastructure, software, models and robotics.

Agentic AI — Logo Crawl

Next Steps

Ready to Get Started?

Build and scale secure AI agents with NVIDIA Open Agent Safety Platform.

Try OpenShell

Build

Access the open source OpenShell repository on GitHub.

Start Building

Documentation

Learn more about OpenShell by exploring the documentation.

Learn More

FAQ

Each component serves a distinct role:

NVIDIA OpenShell governs how agents execute, what they can access and change, and where inference runs.
NVIDIA Sentry, powered by NVIDIA DOCA and running on NVIDIA BlueField-4, provides continuous, tamper-proof, out-of-band observability into agent behavior and enforces security policies in real time at line speed.
NVIDIA Vera CPUs provide high-performance, power-efficient compute for orchestration, sandboxed code execution, and data processing.

Learn more about NVIDIA Cybersecurity and NVIDIA AI Security Research.

Prompts, model safeguards, and agent frameworks influence what an agent attempts to do. Runtime controls enforce what it is allowed to do. OpenShell applies policy outside the agent process, while NVIDIA Sentry with NVIDIA BlueField-4 adds an independent security layer outside agent and host software.

Yes. OpenShell supports agents such as Claude Code, Codex, OpenCode, GitHub Copilot CLI, and OpenClaw. Teams can also bring custom agents and sandbox images. OpenShell supports open and closed models and provides a common runtime policy layer across agent workflows.

No. OpenShell can run on supported local, on-premises, cloud, and Kubernetes infrastructure without BlueField-4. On systems with BlueField-4, NVIDIA Sentry adds hardware-isolated monitoring and policy enforcement that remain operational if the host or workload is compromised.

OpenShell provides an audit trail of allow and deny decisions and supports centralized collection of sandbox logs. Agents can request policy changes, with optional automatic approvals constrained by approved policy limits and enterprise governance. Operators retain control over the boundaries agents must follow.