Box adds Nvidia Nemotron to agent-focused search

Announced at Box’s developer conference on Thursday, Content Retrieval uses metadata generated by Nemotron to improve how agents respond to queries. Box says the approach should make searches faster, more accurate and less expensive. Nvidia vice president of generative AI software Kari Briski told TechCrunch she expects answers to arrive four times faster, with 20% fewer errors and 90% fewer tokens used.

Box CEO Aaron Levie said the system can return the document sections most relevant to a user’s question or an agent’s task, rather than making the agent read each document from start to finish. The feature is based on Nvidia’s Nemotron Embed project, which provides open-weight embedding models for automated retrieval. Nvidia said a Nemotron release in July topped the RTEB benchmark, which measures retrieval quality.

Box says the tool may be particularly useful with active agents. Lawyers, for example, could use it to find relevant clauses and precedents. Because embeddings can map concepts as well as keywords, searches may surface related legal themes even when the wording differs.

Although Nemotron models are freely available, the system also depends on Box’s existing metadata and permission controls, which could make it difficult to reproduce from scratch. Levie told TechCrunch the product combines “five different technologies into one simple API.”