JetPack 7.2: Jetson Software Goes Agentic with Jetson Linux 39.2

Robotics & Edge Computing Jetson Systems Jetson AGX Orin
, , , ,
1

NVIDIA JetPack 7.2 is now live, bringing an agentic-ready software foundation to Jetson Orin and Jetson Thor. JetPack 7.2 packages Jetson Linux 39.2 with Linux kernel 6.8 and an Ubuntu 24.04-based root file system, while extending the CUDA 13-based compute stack across the Jetson portfolio.

As AI agents move from the digital world into the physical world, JetPack 7.2 gives developers a production-grade stack for building agentic systems across robotics, industrial automation, inspection, vision AI, and edge AI. Jetson is now NemoClaw-ready, supports one-command NemoClaw setup, and introduces Jetson Agentic Skills that help agents build, configure, optimize, and measure Jetson software stacks.

Agentic AI on Jetson is built around three layers:

  • JetPack 7.2 (BSP): operating system, compute stack, determinism, and production-ready Jetson Linux.

  • Jetson agent skills: repeatable, agent-executable workflows for Jetson Linux customization, memory optimization, model benchmarking, and deployment configuration.

  • NemoClaw (Agent Orchestrator): a single-command path to deploy agentic AI workflows on Jetson.

The layers of JetPack 7.2 stack explained in detail in the figure below.

image
image2048×1037 169 KB

Jetson Agent Skills empower developers to rapidly customize Linux platforms, optimize memory footprints, and benchmark AI models through automated, agent-driven workflows that reduce time-to-market and improve deployment efficiency. Visit the Skills GitHub to start using Jetson Agent Skills, Jetson device-side skills and Jetson BSP skills.

image
image1018×490 38 KB

Read more about this release in the latest NVIDIA blog, Deploy Agentic-Ready AI at the Edge with Memory Efficiency in NVIDIA JetPack 7.2

What’s new in JetPack 7.2

NOTE: For full details, refer to the Jetson Linux 39.2 release notes.

  • Adds support for the Jetson Orin product family within JetPack 7 releases, bringing Jetson Orin and Jetson Thor onto a unified JetPack 7 foundation.

  • Makes Jetson NemoClaw-ready out of the box, with single-command NemoClaw setup: curl -fsSL https://www.nvidia.com/nemoclaw.sh | bash

  • Introduces Jetson Agent Skills for automating Jetson development workflows:

    • Jetson Linux customization skills help agents build and customize BSPs for custom carrier boards, including I/O, clocks, fan control, and power profiles.

    • Memory optimization skills help agents tune bootloader carveouts, kernel memory reservations, user-space services, and inference memory usage to reduce TCO and run larger workloads on smaller memory footprints.

    • Model benchmarking skills help agents identify the best model, runtime, and optimization path for a target Jetson device and use case.

  • Extends the Ubuntu 24.04, Linux kernel 6.8, and CUDA 13-based compute stack across Jetson Orin and Jetson Thor.

  • Enables Multi-Instance GPU (MIG) on Jetson Thor T5000 as a technology preview feature.

  • Adds Jetson AGX Orin 32GB Super Mode (MAXN_SUPER), increasing AI performance from 200 TOPS to 241 TOPS.

  • Introduces a unified ISO-based installation method for both Jetson Orin and Jetson Thor developer kits.

  • Provides official NVIDIA Open Embedded/Yocto recipes for Jetson Orin and Jetson Thor platforms through the OE4T Yocto Project for Jetson.

  • Includes Jetson SIPL API Package v2.0.0, expanding SIPL camera support and introducing a unified camera framework for GMSL and CoE use cases.

JetPack 7.2 supported components

  • Jetson Linux 39.2

  • CUDA 13.2.1

  • cuDNN 9.20.0

  • TensorRT 10.16.2

  • VPI 4.1.3

  • Vulkan 1.4

  • OpenGL 4.6

  • OpenGLES 3.2

  • DeepStream 8.0

  • Holoscan SDK 3.9.0

  • Nsight Systems 2026.2

  • Nsight Graphics 2025.3

  • NVIDIA Container Toolkit 1.19

Installation

You can install JetPack 7.2 using any of the methods below:

Once Jetson Linux is flashed, you can install the compute stack using SDK Manager on a Linux host or by running the following on Jetson:

"sudo apt update

sudo apt install nvidia-jetpack"

Important notes

  • There is no SD card image for Jetson Orin Nano Developer Kit in this release. Use the unified ISO image to flash the developer kit using a USB stick.

  • Manual flashing instructions have changed for Jetson Thor because of the SBSA architecture. Please follow the instructions in the flashing section in Jetson Linux Developer Guide carefully.

  • MIG support on Jetson Thor T5000 is a technology preview feature.

  • For known issues, migration details, and implementation notes, please review the Jetson Linux 39.2 release notes.

Explore AI Models & Containers

Please visit the Jetson AI Lab to explore the most popular open-source AI models, optimized and ready to run on Jetson. For NVIDIA-supported containers and software stacks, visit the NGC Catalog, which provides a wide range of validated containers for Jetson platforms.

The Jetson Yocto Ecosystem

With official Yocto Project support on Jetson, NVIDIA is expanding the Jetson ecosystem to help developers accelerate and simplify production development.

Distribution and ISV partners include Konsulko Group and Peridio, offering complete OS solutions and Balena, providing container-based fleet management and deployment at scale.

Engineering and services partners including Neurealm, RidgeRun, and WindRiver provide embedded Linux expertise, BSP customization, multimedia pipeline development, NRE services, and long-term platform support.

Hardware ecosystem partners including AAEON, Advantech, Antmicro, ASUS, AVerMedia, Connect Tech, EDOM, and YUAN provide Jetson-optimized carrier boards, edge AI systems, industrial embedded platforms, video capture solutions, and reference designs.

JetPack 7.2 resources