Agentic AI
Box embeds Nvidia's Nemotron to power smarter agent-based document search
The content management company is integrating Nvidia's embedding technology to help AI agents retrieve and analyze documents faster and with fewer errors.
TechCrunch — AI
Box unveiled a new retrieval system at its developer conference that brings Nvidia's Nemotron embedding technology directly into its platform, enabling AI agents to locate and process information more efficiently.
A smarter approach to search#
Box Content Retrieval lets agents tap into rich semantic metadata generated by Nemotron embeddings rather than simply scanning raw documents word-by-word. The system pairs agents with deeper contextual understanding, allowing them to pull the most relevant document chunks in response to user queries or tasks.
According to Nvidia's VP of generative AI software, Kari Briski, the partnership delivers measurable gains: queries run roughly four times faster, error rates drop by 20%, and token consumption falls by 90%.
How it works for real users#
The practical advantage shines in knowledge-intensive fields. Box gave lawyers as an example—they can now locate relevant contract clauses and legal precedents far more readily because Nemotron's embeddings capture conceptual relationships, not just keyword matches. If a contract uses different language than a precedent, the system can still surface the connection.
"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," Box CEO Aaron Levie explained to TechCrunch.
Why this matters#
While Nvidia's Nemotron models are open-weight and freely available, Box's integration packs additional layers of value—pre-existing metadata schemas, permission controls, and platform-specific optimizations. "We're probably packing in five different technologies into one simple API," Levie said. This combination is hard to replicate independently.
Nemotron itself has gained traction in the retrieval space; Nvidia's July release ranked first on the RTEB benchmark, a standard measure of embedding quality. Major AI labs have historically deprioritized embedding models, despite their critical role in helping agents pre-process and organize data.
Rollout timeline#
The feature will become available to Box Enterprise Advance customers beginning early 2027.

