The key facts

  • Box Content Retrieval uses Nvidia Nemotron embeddings for metadata-based search
  • Nvidia expects four times faster queries, 20% fewer errors and 90% fewer tokens
  • The feature will become available to Enterprise Advance customers early next year

A new search system for agents

At its developer conference on Thursday, Box introduced a feature called Box Content Retrieval, which embeds Nvidia's Nemotron embedding system into the core of the product. The feature allows AI agents to answer queries more effectively by relying on metadata generated by Nemotron. According to Nvidia's vice president of generative AI software Kari Briski, queries will run four times faster, with 20% fewer errors and 90% fewer tokens consumed.

How it works and who it's for

The feature implements Nvidia's Nemotron Embed project — open embedding models designed to improve automated retrieval systems. In the July Nemotron release, such models topped the RTEB benchmark, which measures retrieval quality, as also noted in Nvidia's blog post on Hugging Face. Box CEO Aaron Levie explained that instead of an agent reading every document, the system returns the most relevant fragments:

Availability and a barrier for competitors

The feature will be available to all Box Enterprise Advance customers, with deployment at the beginning of next year. Although the Nemotron models are freely available, the system relies heavily on Box's existing metadata and access rights management mechanisms, which makes it difficult to replicate from scratch.