NVIDIA DeepStream 9.1 is now open source: 13 new AI tools

NVIDIA released NVIDIA DeepStream 9.1 today, rolling out an open-source update to its video analytics SDK with 13 new agentic skills, Multi-View 3D Tracking, AutoMagicCalib, and support for JetPack 7.2, according to the company.

The headline feature is the shift to natural-language prompts for building, managing, and changing video analytics pipelines. Instead of wiring every stage by hand, you can describe what you want in plain language. NVIDIA says work that used to take weeks can now be done in hours.

DeepStream 9.1 is also moving to open source through NVIDIA’s unified GitHub repository. That should make adoption simpler and make it easier to keep up with updates.

Multi-View 3D Tracking, or MV3DT, gives multiple cameras a shared 3D coordinate system, so a person, cart, or vehicle keeps the same identity as it moves from one view to another. NVIDIA says that matters for things like warehouse safety, retail traffic measurement, and smart-building monitoring.

AutoMagicCalib tackles another headache. According to NVIDIA, it automates one of the most frustrating setup steps in multi-camera deployments.

If you build security, logistics, operations, or monitoring systems, this looks like a release to pay attention to. Still, independent real-world testing of the agentic tools is limited so far, and the usual problems around infrastructure, data quality, security, governance, and older-system integration aren’t disappearing overnight.

It also lines up with NVIDIA’s broader Metropolis push, from data center hardware all the way down to NVIDIA Jetson Orin and NVIDIA Thor edge devices.

You can download NVIDIA DeepStream 9.1 as an open-source release from NVIDIA’s unified GitHub repository.