A couple weeks ago, the Peridio team was together at our Nashville HQ for our exec offsite. While the rest of us were locked in on roadmap and strategy, my co-founder Justin took over a corner of the office and went into what he calls "infinity mode" — two laptops, three Jetson dev kits, screens stacked across his desk, the full Orin Nano / AGX Orin / Thor lineup wired in and running. OS images compiling at a pace that would make most embedded engineers think it was fake. He was rehearsing the technical walkthrough we'd give NVIDIA's Jetson team a couple days later. Watching it come together, I had a thought I've had a few times this year: this is the moment we built Peridio for.
Today, with the launch of NVIDIA Jetpack 7.2 that moment is here.
Agentic AI is moving from the data center to the physical world. JetPack 7.2 is the release that makes it a production reality — and we're honored to be included as a launch partner.
JetPack 7.2 is a watershed release. I don't say that lightly. Three things ship in this release that together change what's possible on Jetson.
For years, the Yocto-on-Jetson conversation lived in community efforts and partner work. With 7.2, NVIDIA is shipping Yocto as a first-class citizen alongside the L4T/JetPack Ubuntu reference. This is a serious signal: NVIDIA is meeting the production-grade Linux demand head-on.
A new middle layer of agentic skills: Linux customization, memory optimization, model benchmarking, built directly from NVIDIA's documentation and design guides. The agent-native developer experience just landed on Jetson, and tasks that used to take weeks resolve in days.
NVIDIA's agentic AI framework now deploys to production-grade Jetson with a single command. Physical AI agents are no longer a demo. They're a deployment target.
CUDA 13 on Orin, MIG plus the real-time kernel on Thor, an AGX Orin 32GB performance bump to 241 TOPS — all material. But the three layers above are what change the trajectory of physical AI deployments over the next eighteen months.
The shorthand I've been using lately: NVIDIA is at least 4 years ahead, and it's not even close. JetPack got millions of developers building. The Jetson platform is in deployment across robotics, industrial inspection, agriculture, healthcare, humanoids — fields where physical AI is shifting from pilots to fleets.
But every team that scales past a few dozen devices runs into the same wall: a development-optimized OS isn't a production-optimized OS. The general-purpose Ubuntu that gets you to a working prototype isn't the right foundation for a thousand devices running unattended for ten years in regulated environments. That's not a critique of JetPack, it's a design boundary. JetPack was built to maximize the developer on-ramp, and it's the best one in the market. The production layer is a different job.
That's exactly what JetPack 7.2's Yocto support addresses. By giving production teams a path to a leaner, reproducible, minimal-attack-surface Linux foundation, NVIDIA has closed the loop between development and production on the same platform. Engineers no longer have to choose between Jetson's ecosystem and the OS rigor their deployment demands.
We started Peridio because we believed the production layer for physical AI should be a product, not an afterthought.
Avocado OS is a turnkey, immutable, Yocto-based Linux distribution for NVIDIA Jetson. We build Yocto so your team doesn't have to. You get the production properties: reproducible builds, minimal footprint, secure boot, full-disk encryption, atomic OTA with rollback, without the staffing burden that has historically gated Yocto adoption.
Peridio Core is the management plane: OTA with one-command rollback, fleet observability, remote access, CVE management, and long-term support on a defined SLA.
Together, they let teams move from a Jetson dev kit to a 10,000-device fleet without rebuilding the OS layer or staffing a Yocto team along the way. Avocado OS supports the full Jetson family from a single YAML config — Orin Nano, Orin NX, AGX Orin, and Thor — and NemoClaw runs on top of it as a first-class deployment target.
We've been an NVIDIA Preferred Partner and a Yocto Project member for some time. Being included as a launch partner in JetPack 7.2 is the most concrete expression yet of where we believe the Jetson production story is headed.
Alongside JetPack 7.2, we're closing in on our official Avocado OS 1.0 release as a fast follow. Without spoiling the launch entirely, a few things engineers in the Jetson ecosystem can expect:
A radically different developer on-ramp. We've been quietly building something that's going to change how developers approach Jetson — across more host platforms than the Linux-only audience the ecosystem has historically reached. If you've ever had an NVIDIA-stickered laptop you weren't actually able to build on, this one's for you.
Agent-native development. Avocado has been designed from the ground up for a world where coding agents help you compose runtimes, find packages, and debug devices on real hardware. The 1.0 release brings that experience to life — and we think it'll feel like a step-change.
Hardware-in-the-loop, accelerated. Compile on your host, see your changes on the device in seconds. We've been working on the underlying infrastructure to make this dramatically faster than anything Yocto's traditional NFS path has historically offered.
We'll have more to share when 1.0 lands. If you want a heads-up the day it ships, join our discord — or follow along at docs.peridio.com, where we'll be opening free developer accounts as 1.0 goes live.
When we started Peridio, we made an early bet on Yocto as the production foundation for physical AI. We did it because we didn't want to put our customers in an Apple-versus-Microsoft moment between development velocity and production rigor. We wanted to give engineers the developer experience of JetPack with the production properties of Yocto, in one tool, from one vendor.
JetPack 7.2 is NVIDIA shipping the foundation that bet was built on. The fact that Yocto is now a first-class part of the Jetson story is the strongest possible signal that the production-AI moment has arrived.
We are a small team, building a serious platform, alongside one of the most important compute companies of our generation. That's a privilege we don't take lightly.
To Shashank, Ilies, Thejas, Kathy, and the broader Jetson team at NVIDIA — thank you. The work you all are doing is going to make physical AI real, and we couldn't be more excited to build alongside you.
If you're already on Jetson and thinking about the move to production, this is the moment. The teams that get their OS strategy right in the next twelve months will be the ones who scale cleanly from dozens to thousands of devices. The teams that don't will spend 2027 unwinding decisions they made when shipping felt far away.
Engineers → Get started with Avocado OS on Jetson — free developer account, up and running in minutes, not weeks.
Engineering and product leaders → Our new comprehensive guide with Advantech, Shipping Vision AI on NVIDIA Jetson, walks through the prototype-to-production path on Jetson, choosing an OS strategy, and where production teams hit the wall.
Physical AI just got real. Let's go build it.
-Bill