09/16/2026
For the past decade, adding new functionality to a video analytics system meant building entire AI models from scratch. That used to be a lengthy and expensive development cycle, requiring teams to build, train, and validate custom AI models.
Today, we’ve changed that 🔥
Meet StreamVLM™, a Vision-Language Model detection engine that turns a simple plain-language description into a functional video detection. Just input a descriptive criteria (i.e., «detect a person lying on the ground», «alert when graffiti appears on a wall», «identify flooding in the underpass»), and the system instantly begins searching for that specific scenario across selected cameras, without any prior dataset creation or model training ✔️
«Public safety agencies have endless monitoring needs but limited resources to build individual models», says Serge Smirnoff, Head of PR at IREX. «StreamVLM completely transforms how camera networks operate. It’s no longer a data science project, but a sentence typed in by an officer who is familiar with the neighborhood, the station, or the campus».