Box builds Nvidia's Nemotron into a retrieval system for agents

Box showed a feature called Box Content Retrieval at its developer conference on Thursday, TechCrunch reported. It builds Nvidia's Nemotron embedding system into Box so agents can answer from embedded metadata instead of reading each document. Nvidia's vice president of generative AI software, Kari Briski, told TechCrunch she expects queries to be answered four times faster, with 20 percent fewer errors and 90 percent fewer tokens.

TechCrunch said the feature will be available to Box Enterprise Advance customers and is set to roll out early next year. Chief executive Aaron Levie told TechCrunch the change is meant to return the chunks of a document that matter to a query, rather than making an agent read the whole file. The report described the work as an implementation of Nvidia's Nemotron Embed project. TechCrunch said a Nemotron release in July led the RTEB retrieval benchmark, and that Box expects the system to help with tasks such as finding clauses, because the embeddings map concepts and not only keywords. Levie told TechCrunch the product packs several technologies into one API and relies on Box's existing metadata and permission controls.