Box integrates Nvidia's Nemotron embedding model into new Box Content Retrieval system for agents
At its developer conference, Box introduced Box Content Retrieval, a feature combining Nvidia's Nemotron embedding model with Box's metadata and permissions. According to Nvidia, it should deliver 4x faster queries, 20% fewer errors, and 90% fewer tokens. Available to Box Enterprise Advance customers starting…
At its developer conference, Box introduced a new feature called Box Content Retrieval, which embeds Nvidia's Nemotron embedding system into the core of the product. The feature is intended to let agents answer queries more accurately by giving them only the most relevant text passages, identified via embeddings generated by the Nemotron model, instead of having them read entire documents.
According to Kari Briski, Nvidia's VP of generative AI software, queries should be up to four times faster, have 20% fewer errors, and consume 90% fewer tokens. This is a claim made by the company that the source does not independently verify. The feature will be available to all customers with a Box Enterprise Advance subscription, with rollout planned for early 2027.
Technologically, this is an implementation of Nvidia's Nemotron Embed project, which offers open-weight embedding models to improve automated retrieval systems — an area that, according to the source, major labs often overlook. A recent Nemotron version from July tops the RTEB benchmark, which measures retrieval quality. As an example use case, Box cites lawyers who, thanks to the mapping of semantic connections (not just keywords), can more easily find relevant clauses and precedents even when phrasing differs.
Box CEO Aaron Levie said the system combines several technologies into a single API — besides the Nemotron embeddings themselves, this mainly involves Box's existing metadata and permission management tools, which he said makes the solution harder for competitors to easily replicate.
Why it matters
For companies using Box, this is a concrete promise of lower AI query costs and more accurate document search for agents, which is practically useful especially in processes that involve working with large volumes of documents (e.g. legal work). At the same time, it shows that Nvidia's open-weight embedding models, so far less visible in enterprise deployment compared to large LLMs, are finding commercial use through integration into existing enterprise platforms.
Two audiences, two different impacts
What this means
For individuals
Individuals working with documents in Box (according to the source, for example lawyers searching for clauses and precedents) should, according to Box, get more relevant search results, because the system maps semantic connections rather than just keywords.
For a business
Companies with a Box Enterprise Advance subscription will get, starting in early 2027, a tool that, according to Nvidia and Box, should reduce the cost of AI queries (up to 90% fewer tokens) and make document search for enterprise agents both faster and more accurate.
ProductivityCheck the original
Event sources
only one source so far · 1 publisher, 1 independent. We count feeds from the same owner only once.