Box Integrates Nvidia's Nemotron Into New Agent-Focused Content Retrieval Tool

At its developer conference on Thursday, Box introduced a feature that places Nvidia's Nemotron embedding system at the heart of its product.

The tool, Box Content Retrieval, lets agents answer queries more effectively by drawing on embedded metadata generated by Nemotron. According to Box, users can expect faster, more accurate and cheaper queries. Nvidia's VP of generative AI software, Kari Briski, told TechCrunch she expects answers four times faster, with 20% fewer errors and 90% fewer tokens used.

The system will be offered to all of Box's Enterprise Advance customers, with rollout beginning early next year.

Box CEO Aaron Levie said the feature sharply improves how relevant search is for agents. Rather than having an agent read through every document, he said, the system can return chunks from the documents most relevant to a user's query or the agent's task.

The feature implements Nvidia's Nemotron Embed project, which provides open-weight embedding models for improving automated retrieval. Such models have often been overlooked by major labs, though the area has been a focus for Nvidia's open-source work; its July Nemotron release topped the RTEB benchmark, which measures retrieval quality.

Box says the system is most useful alongside active agents, citing lawyers who could more easily locate relevant clauses and precedents. Because Nemotron's embeddings map underlying concepts rather than simple keywords, searches can surface related legal themes even when the wording or language differs.

While the Nemotron models are freely available, the system leans heavily on Box's existing metadata and permission controls, making it hard to recreate from scratch. Levie told TechCrunch the company is "probably packing in five different technologies into one simple API."

Source: TechCrunch