Box has expanded its partnership with Nvidia to rebuild how AI agents find answers inside company files. The centrepiece is Box Context Retrieval, a new search engine built on Nvidia’s Nemotron models that turns enterprise documents into a structured knowledge layer for agents. Box says it can lift search accuracy by 20 percent, cut token use by 90 percent, and answer questions four times faster. The engine reaches general availability for Enterprise Advanced customers early next year, alongside a new workflow builder and stronger threat protection.
## What Box announced
The announcement came at Box’s developer conference on Thursday, where chief executive Aaron Levie shared the stage with Nvidia chief executive Jensen Huang. Box Context Retrieval is the headline act: instead of an agent reading through every document, the engine pre-processes company files with Nemotron embedding models and returns only the passages that matter for the task at hand.
Box also unveiled Box Architect, a tool that lets teams build automated workflows by describing a process in plain language. It begins rolling out in beta to Enterprise Advanced customers in the fourth quarter. New threat protection against bulk downloads, account takeovers and mass deletion will also arrive early next year for Enterprise Advanced and Shield customers.
## How the retrieval engine works
The engine pairs keyword matching with semantic search powered by Nvidia’s Nemotron Retriever models. Nemotron’s embeddings map the concepts inside a document rather than its words, so an agent can surface legal themes even when contracts use different wording or language. Every request checks Box permissions, so nobody sees files they are not allowed to open.
The numbers Box and Nvidia quote are projections, not independent test results. Nvidia’s generative AI software chief, Kari Briski, told TechCrunch she expects queries answered four times faster, with 20 percent fewer errors and 90 percent fewer tokens used. Box says the Nemotron models it uses topped a retrieval-quality benchmark when they were released in July.
## Our read
This is the part of the agent race that gets the least attention and may matter the most. Models keep getting smarter, but most [enterprise agents fail](https://geroundtable.com/google-agents-exam-cannot-pass/) at the unglamorous step: finding the right passage in the right file, with the right permissions, before they start reasoning. Box is betting that retrieval, not another model, is the bottleneck worth owning.
Nvidia’s angle is just as telling. The chipmaker keeps widening its open Nemotron family — embeddings, rerankers, parsers — and placing them inside other companies’ products. Every deployment like Box’s makes Nemotron the default plumbing for enterprise AI, and each one still runs best on [Nvidia hardware](https://geroundtable.com/nvidia-vera-cpu-agent-benchmarks/). It is a software moat built on open weights.
The honest caveat is timing. The retrieval engine does not reach general availability until early next year, and its headline figures are company projections that no independent test has confirmed. Box’s real test comes when Enterprise Advanced customers put it to work on messy, real-world file stores — and when rivals with their own retrieval layers answer.
## Why It Matters
Enterprise AI has a retrieval problem, and retrieval is where most agent projects stall. If Box’s engine delivers even part of what it promises — faster answers, far fewer tokens, permission-aware results — it lowers the cost of [every agent a company runs](https://geroundtable.com/claude-enterprise-app-store/) on its files. That kind of win shows up in budgets before it shows up in headlines.
## Background
Box and Nvidia are not new partners. Box already offers Nvidia’s open-weight Nemotron models inside Box AI, and its July Nemotron release topped a retrieval benchmark. The new engine turns that partnership into a product: Box supplies the files, the permissions and the metadata, while Nvidia supplies the models that make sense of them.
Source: [Reuters report](https://www.reuters.com/business/box-expands-nvidia-partnership-unveils-ai-workflow-builder-2026-11-05/) and [TechCrunch coverage](https://techcrunch.com/2026/11/05/box-taps-nvidias-nemotron-for-new-agent-first-content-retrieval-system/)
0 Comments