Box showed off a new feature, Box Content Retrieval, at its developer conference that builds Nvidia’s Nemotron embedding system into the core of its product. The feature lets agents answer queries by relying on embedded metadata created by Nemotron, rather than reading through entire documents. Box says the result is faster, more accurate, and less expensive queries. Nvidia’s VP of generative AI software, Kari Briski, told TechCrunch she expects answers to be four times faster, with 20% fewer errors and 90% fewer tokens used. The system will roll out early next year to all of Box’s Enterprise Advance customers.

The feature is an implementation of Nvidia’s Nemotron Embed project, which offers open-weight embedding models to improve automated retrieval systems. The article notes that major labs have often overlooked such models, which use LLMs to pre-process data so it can be better used by agents. This area has been a focus for Nvidia’s open-source model project; its recent Nemotron release in July topped the RTEB benchmark, which measures retrieval quality.

Box says the system will be most useful when paired with active agents, citing lawyers as an example: they could find relevant clauses and precedents more easily. Because Nemotron embeddings go beyond simple keywords to map underlying concepts, the search tool can surface related legal themes even when different wording or language is used. While the Nemotron models are freely available, the system also relies heavily on Box’s existing metadata and permission controls, which Box says makes it challenging to recreate from scratch. As CEO Aaron Levie put it, “We’re probably packing in five different technologies into one simple API.”