Box introduced Box Content Retrieval at its developer conference, embedding Nvidia’s Nemotron into its core platform and planning an early-2027 rollout for Enterprise Advance customers.
Nvidia said the system can answer queries four times faster, cut errors by 20% and use 90% fewer tokens by relying on embedded metadata instead of having agents read entire documents.
Box CEO Aaron Levie said the feature returns the most relevant document chunks for an agent’s task, aiming to improve search relevance and lower query costs.
Lawyers were cited as a key use case because Nemotron maps underlying concepts rather than keywords, helping surface related clauses and precedents even when wording differs.
The offering combines Nvidia’s open-weight Nemotron Embed models with Box’s existing metadata and permission controls, which Box says makes the system hard to replicate from scratch.