Box expands Nvidia partnership with AI retrieval and workflow tools

Box announced new AI retrieval and workflow products with Nvidia on November 5, 2026, including a search engine intended to help AI tools find answers in company files. Reuters reported that the announcement expands the companies’ partnership; the products are aimed at Box business customers, with availability planned in stages.

How Box Context Retrieval is designed to work

Reuters called the search product Box Context Retrieval and said it is built on Nvidia’s Nemotron models. TechCrunch referred to it as Box Content Retrieval, describing a system that uses metadata from Nemotron embeddings to return document passages relevant to an agent’s query or task. The reports use different names for the feature; Box’s CEO, Aaron Levie, told TechCrunch that the approach can return relevant document chunks instead of making an agent read each document in full.

Rollout plans and performance claims

Box said Context Retrieval would become generally available to Enterprise Advanced customers early in 2027, according to Reuters. Box also said the system could improve search accuracy by 20 per cent, reduce token use by 90 per cent and make question-answering four times faster. TechCrunch separately attributed estimates of four-times-faster responses, 20 per cent fewer errors and 90 per cent fewer tokens to Nvidia generative AI software vice-president Kari Briski. The accounts describe the accuracy measure differently, and neither supplies independent test results or the methods behind the estimates.

Box Architect, the second announced product, lets teams describe a process in plain language to create AI-driven workflows. Reuters said its beta rollout to Enterprise Advanced customers was planned for the fourth quarter, without specifying the year. Box also announced protection against bulk downloads, account takeovers and mass deletion, planned for Enterprise Advanced and Shield customers early next year.