Box has announced a new content retrieval feature that integrates Nvidia’s Nemotron embedding system to enhance the performance of SI agents. Unveiled at the company’s developer conference on Thursday, the update aims to make query responses faster, more accurate, and less expensive for enterprise users.
What Happened
The new feature, called Box Content Retrieval, embeds metadata created by Nvidia’s Nemotron system into the core of Box’s product. This allows SI agents to answer queries more effectively by leveraging this pre-processed data. Kari Briski, Nvidia’s VP of generative SI software, told TechCrunch that the company expects queries to be answered four times faster, with 20% fewer errors and 90% fewer tokens used. The feature will be available to all Box Enterprise Advance customers, with a rollout scheduled for early next year.
Box CEO Aaron Levie stated that the integration dramatically boosts the relevance of search technology for agents. "Instead of an agent having to kind of read through each document, we can send back chunks from the documents that are most relevant to the user’s query or whatever the task is that the agent is doing," Levie said. The system is an implementation of Nvidia’s Nemotron Embed project, which provides open-weight embedding models to improve automated retrieval systems.
Why It Matters
This move highlights the growing importance of embedding models in agentic SI workflows. While major labs have often overlooked these models, which use LLMs to pre-process data for better agent utilization, Nvidia has focused on this sector with its open-source model project. The recent Nemotron release topped the RTEB benchmark in July, measuring retrieval quality. For developers and enterprises, the shift means SI agents can access specific, relevant document chunks rather than processing entire files, reducing computational overhead.
The system is particularly useful when paired with active SI agents, such as those used by lawyers to find relevant clauses and precedents. Because Nemotron’s embeddings map underlying concepts beyond simple keywords, the search tool can surface related legal themes even when different wording or language is used. However, recreating this capability from scratch is challenging, as it relies heavily on Box’s pre-existing metadata and permission controls. "We’re probably packing in five different technologies into one simple API," Levie noted.
The Bottom Line
Box’s integration of Nvidia’s Nemotron embeddings marks a significant step in optimizing SI agent performance for enterprise content retrieval. By focusing on relevant data chunks and leveraging advanced embedding technology, the update promises faster and more accurate query responses, setting a new standard for agentic SI tools in the enterprise sector.