Nvidia just dropped a massive open-source toolkit of over 80 AI agent skills and tools under its Metropolis framework, aimed squarely at developers building vision AI for manufacturing, infrastructure, robotics, and industrial digital twins. The release landed on June 1 and represents one of the company’s most comprehensive pushes yet to make physical AI accessible to enterprises that want to turn raw video feeds into real-time operational intelligence.
What Nvidia actually built
The Metropolis platform now includes several key components designed to lower the barrier for building sophisticated vision AI agents. VSS Blueprint 3.2 enhances video search and summarization capabilities. DeepStream 9.1 handles video processing at scale. TAO 7 is focused on accelerating vision AI model development.
The toolkit supports edge-to-cloud deployment, meaning these AI agents can run on devices at the physical location (a camera on a warehouse wall, for instance) or scale up to cloud infrastructure when heavier processing is needed. Nvidia integrated this with its Cosmos and Omniverse technologies, enabling natural language-driven workflows across simulated and real-world environments.
Developers can now create customizable AI agents using natural language prompts rather than writing complex pipelines from scratch, significantly reducing time and complexity in transforming raw video data into actionable insights.
