At its developer conference on Thursday, Box demonstrated a new capability called Box Content Retrieval that embeds Nvidia’s Nemotron embedding system directly into its product core. The feature is designed so agents and search systems can rely on Nemotron-generated embedded metadata to respond to queries more effectively.
What Box says it delivers
Box positions the feature as producing faster, more accurate, and less expensive queries. Kari Briski, Nvidia’s vice president of Generative AI software, told TechCrunch she expects queries to be answered four times faster, with 20% fewer errors and 90% fewer tokens used when leveraging the integration.
The company said the system will be available to all Enterprise Advance customers, with rollout beginning early next year.
How it works in practice
Box CEO Aaron Levie said the approach significantly improves relevance for agents. Rather than having an agent read through full documents, the system can return document chunks that are most relevant to the user’s query or the task the agent is performing.
The feature implements Nvidia’s Nemotron Embed project, which provides open-weight embedding models to enhance automated retrieval systems. While major labs have often overlooked embedding models, Nvidia’s open-source effort has focused on them; Nemotron’s most recent July release ranked first on the RTEB benchmark, which measures retrieval quality.
Use cases and limitations
Box expects the system to be most useful when paired with more active agents that can operate fluidly over Nemotron-processed data. The company gave the example of lawyers, who would be able to locate relevant clauses and precedents more easily because Nemotron embeddings map underlying concepts beyond simple keyword matches. That lets a search tool surface related legal concepts even when different terminology is used.
Although the Nemotron models themselves are freely available, Box’s implementation depends heavily on the company’s existing metadata and permission controls, making the full solution difficult to recreate from scratch. As Levie told TechCrunch: “we’re probably packing in five different technologies into one simple API.”
In short, Box Content Retrieval combines Nvidia’s Nemotron embeddings with Box’s metadata and access-control layers to deliver more concept-aware retrieval that the company says will be faster, more accurate, and more cost-efficient for enterprise customers.
