back

JetPatch Unveils Enterprise Guardrails for NVIDIA NemoClaw, Bringing Control to the Autonomous Agent Frontier

Mar 18, 2026
 

JetPatch Unveils Enterprise Guardrails for NVIDIA NemoClaw, Bringing Control to the Autonomous Agent Frontier

PR Newswire (New York, March 18, 2026) – An enterprise control plane for OpenShell that enables organizations to deploy, throttle, and hard-stop autonomous AI agents in real time.

JetPatch, a leader in automated infrastructure governance, today announced the availability of its Enterprise Control Plane for NVIDIA NemoClaw and OpenShell. Introduced in the wake of NVIDIA’s GTC 2026 announcement, JetPatch delivers the operational layer enterprises need to turn autonomous “Claws” into governed corporate assets.

The Cockpit for Autonomous AI Agents

As AI shifts from passive chatbots to proactive agents capable of writing code, spawning sub-agents, and acting independently, the risk of execution drift has become a growing priority for enterprise IT and security teams. While NVIDIA OpenShell provides the secure sandbox, JetPatch provides the operational control plane.

Governor-Level Control for Autonomous Fleets

JetPatch transforms the complex policy.yaml management of OpenShell into a scalable enterprise workflow. IT operations teams can:

  • Kill-Switch Integration: Instantly pause or suspend autonomous agent processes across global infrastructure when anomalous behavior is detected.
  • Token and CPU Throttling: Prevent agent runaway and cost spikes by dynamically limiting token consumption and compute utilization.
  • One-Click Claw Deployment: Automate deployment of the NemoClaw stack across hybrid cloud and RTX-powered edge environments.
  • Declarative Policy Governance: Centrally manage and enforce OpenShell security sandboxes so agents cannot operate outside defined inference boundaries.

“The era of the autonomous agent has arrived, but it cannot survive in the enterprise without a leash,” said Ronald Ranaldi, Chief Revenue Officer at JetPatch. “NVIDIA NemoClaw is the engine; JetPatch is the cockpit. We are giving organizations the confidence to let their ‘Claws’ run 24/7, knowing they have total visibility and a hard-stop safety switch at their fingertips.”

Seamless Ecosystem Integration

JetPatch bridges the gap between AI innovation and traditional enterprise security. The platform works alongside existing security platforms, including CrowdStrike Falcon, helping organizations connect autonomous agent operations with broader enterprise security workflows.

JetPatch for NemoClaw is available today for early-access customers across multiple verticals, including financial services, healthcare, and government.

Update (October 2026): One Console for Every Agent, Not Only AI Agents

The NemoClaw controls described above run in JetPatch Agent Manager, the same console that has managed server tool agents in production for over a decade. Enterprise servers typically run 8 to 15 software agents for monitoring, security, backup, performance and compliance. JetPatch has 45+ pre-built packages for them, including Splunk, CrowdStrike, Rapid7 InsightVM and Zabbix, can package any agent with an unattended installer, alongside AI agents such as NVIDIA NemoClaw, OpenShell, Claude-based and CrewAI agents.

Both kinds of agent get the same controls, the same policies and the same audit trail:

  • Deploy: roll agents out to thousands of servers across cloud, on-premises and hybrid infrastructure, with policies bound at startup.
  • Update: upgrade agent versions fleet-wide with zero-touch automation.
  • Start, stop and kill switch: stop any agent with one action, or pause agent processes across the estate when behavior looks wrong.
  • Throttle: cap CPU and memory per agent so tools never slow down the servers they run on, and enforce token budgets for AI agents.
  • Guardrails: run each agent isolated in its own container and allow or block capabilities in real time; Execwall checks every shell command, network call and tool invocation before it reaches the operating system.
  • Self-heal: detect crashed, drifted or silent agents and redeploy, reconfigure or restart them automatically.
  • Audit: one compliance dashboard across every managed agent, with audit trails for SOC 2 and EU AI Act reviews.

Why one pane of glass matters

  • One inventory. Security and audit teams can see every agent on every server, AI or not, in one place. You cannot govern an agent you cannot see.
  • One policy model. AI agents follow the same groups, approvals and rules as the tools you already trust, instead of becoming a separate, less-governed silo.
  • Faster response. When something goes wrong, one kill switch and one audit trail shorten the path from detection to a stopped process, whichever agent caused it.
  • Shared servers, shared resources. AI agents run next to security, monitoring and backup agents. Seeing and throttling them together prevents one agent from starving the others.
  • Less to run. One console to secure, staff and integrate with your ITSM and security tools, instead of a new console for every agent vendor.

Learn more about server and AI agent management and the AI Agent Cockpit and Execwall.

See It in Action

JetPatch ExecWall applies the same principle at the operating system layer: every shell execution, network call, and file operation an AI agent attempts is evaluated against policy before it runs. We recreated CVE-2026-2256, a real indirect prompt-injection attack, end to end. Watch the attack and see how it’s stopped on LinkedIn.

Read the full release on PR Newswire, or contact us to join the early-access program.

Trademark Disclaimer: NVIDIA, NemoClaw, OpenShell, and related marks are trademarks or registered trademarks of NVIDIA Corporation. CrowdStrike, Falcon, and related marks are trademarks or registered trademarks of CrowdStrike, Inc. All other company names, product names, and marks are the property of their respective owners.