Versos launches NVIDIA NeMo agents for video training

Versos AI has launched agents built with NVIDIA NeMo for video training data, aimed at AI teams working with premium video libraries.

The product lets users describe the video dataset they need in plain language. The agents then search Versos' scene-level and frame-level video intelligence, assess footage against the request, and assemble matching material into a structured training dataset.

Versos focuses on licensing video content for AI training. Its latest launch targets a stage of data preparation that can involve manual searching, asset review, and dataset assembly when requests call for specific subjects, actions, and environments.

Users can submit detailed dataset requests covering particular content requirements. The agents identify relevant footage from premium libraries while keeping ownership, licensing, and provenance information attached to the material.

Manual process

That emphasis reflects a wider concern among media owners and AI developers over how training data is sourced and documented. Rights management and provenance tracking have become more prominent as AI companies seek larger, more specialised datasets and content owners push for clearer controls over how their assets are used.

Versos said a single request for video training data can otherwise take days to fulfil through conventional workflows. The company is positioning the agents as a way to reduce that workload by automating parts of search, review, and preparation.

The use of NVIDIA NeMo links the launch to a growing market for software tools that help businesses build AI agents for defined tasks. Here, the task is matching natural-language descriptions of required footage with indexed video content and returning a usable dataset.

Rights focus

The announcement also highlights a commercial issue in the AI training data market: access to premium video libraries with clear licensing terms. For companies developing models that depend on visual data, the ability to trace where material came from and what permissions apply can affect both procurement decisions and compliance processes.

Versos has framed its platform around those requirements by tying content discovery to ownership and usage information. That approach may appeal to rights holders seeking a route into AI licensing without losing visibility into how footage is selected and packaged.

Chris Keevill, co-founder and chief executive officer of Versos AI, was identified by the company as available to discuss how agents are changing video-data curation, what AI teams require from training datasets, and why rights and provenance need to remain part of the workflow.

The launch places Versos in a segment of the AI supply chain between content owners and model developers. As demand for training material becomes more specialised, companies in that position are trying to show that data sourcing can be both faster and more tightly documented than traditional manual methods.

Versos said the agents search its video intelligence at scene and frame level, evaluate footage against user requirements, and assemble selected content into a structured training dataset while keeping ownership, licensing, and provenance information intact.