NVIDIA JetPack 7.2: AI agents on edge devices with a single deployment command

NVIDIA has released JetPack 7.2 — an update to the Jetson platform that simplifies deployment of AI agents on edge devices, reduces memory requirements, and adds native support for the NemoClaw stack with privacy control.

What Changed in JetPack 7.2

JetPack is the base SDK for NVIDIA's entire Jetson lineup: from compact Orin Nano to AGX Orin, which is installed in autonomous vehicles, robots, and industrial equipment. Before version 7.2, developers had to manually configure dozens of dependencies, separately install the runtime, and separately install security tools. Now a single CLI call is sufficient. JetPack 7.2 has embedded native support for deploying NVIDIA NemoClaw, and improved memory optimization allows running large language models that previously required servers with multiple GPUs directly on compact devices.

NemoClaw: Open Security Layer

NemoClaw is an open source layer on top of OpenClaw that adds industrial control tools to the popular agent framework. OpenClaw is good for prototyping, but in scenarios with sensitive data — manufacturing, medicine, defense sector — it lacks built-in protection. NemoClaw fills this gap:

• Access control: the agent uses only explicitly authorized tools and APIs
• Data filtering: sensitive information is processed locally and does not leave the device
• Action audit: every tool call is logged for subsequent verification
• Component isolation: parts of the agent run in separate processes
• Full compatibility with existing OpenClaw agents

For corporate customers this removes the main objection: agents on Jetson devices can now be deployed without risk of data leakage through the cloud.

Agent Skills for Jetson

JetPack 7.2 introduces the concept of NVIDIA agent skills — pre-prepared modules with optimized tools for specific classes of tasks. Skills include working with cameras and sensors in real time, managing robotic manipulators, and local processing of audio and video without transmission to the cloud. This changes the economics of edge AI: instead of expensive infrastructure with cloud backend, the developer gets an autonomous device capable of performing complex agent tasks locally.

This is especially valuable for drones, industrial robots, medical equipment, and IoT — anything where a cloud request is unacceptable due to latencies in the hundreds of milliseconds or strict privacy requirements.

What It Means

NVIDIA is systematically lowering the barrier to entry for edge AI. JetPack 7.2 is not a new chip and not a new architecture, but pragmatic tooling: one CLI call instead of several days of setup, an open security stack instead of homemade solutions, ready-made agent skills instead of writing from scratch. The physical world is preparing to accept a wave of AI agents — and NVIDIA is betting on becoming the platform they run on.