Box has announced a new AI-powered content retrieval system that integrates Nvidia’s Nemotron embedding models to help enterprise AI agents search and analyze documents more efficiently. Introduced at the company’s developer conference on November 5, 2026, Box Content Retrieval is designed to improve search accuracy while reducing processing time and AI-related computing costs.
The feature will become available to Box Enterprise Advanced customers in early 2027. According to Nvidia vice president of generative AI software Kari Briski, the system is expected to deliver search results four times faster, reduce errors by 20%, and consume 90% fewer tokens.
Box Content Retrieval Improves AI Document Search
Box developed the new feature to help AI agents retrieve relevant information from enterprise documents without processing entire files. Instead, the system identifies and returns specific document sections related to a user’s query or an agent’s assigned task.
The technology uses Nvidia’s Nemotron embedding models to represent document content according to its meaning rather than relying exclusively on keyword matching. This allows the retrieval system to identify relevant information even when documents use different terminology or languages.
Box CEO Aaron Levie said the approach improves the relevance of search results for AI agents. By returning selected document passages, the system reduces the amount of information AI models must process when generating responses.
The capability is designed for organizations managing extensive document collections, where searching through individual files can require substantial computing resources.
Nvidia Nemotron Models Support Semantic Search
The integration is based on Nvidia’s Nemotron Embed project, which provides open-weight models for converting text into numerical representations that capture semantic relationships. These embeddings help automated retrieval systems identify information related to a query without requiring exact keyword matches.
Nvidia’s Nemotron embedding models have demonstrated strong performance in retrieval benchmarks, including the RTEB evaluation, which measures the quality of information retrieval systems. A Nemotron model released in July 2026 topped the benchmark, according to Nvidia.
Box identified legal research as one potential application, where AI agents could locate relevant contract clauses, legal precedents, and related documents. The system could retrieve passages addressing similar legal concepts even when their wording differs across documents.
Enterprise Permissions and Metadata Remain Integrated
Although Nvidia distributes the Nemotron models openly, Box Content Retrieval also relies on Box’s existing document metadata and access permission systems. These capabilities help determine which information AI agents can retrieve while maintaining organizational access restrictions.
Levie said the company combines approximately five underlying technologies into a single application programming interface. The integration is intended to simplify development for enterprises building AI agents that need to retrieve information from Box-managed documents.
Box plans to introduce the feature to Enterprise Advanced customers in early 2027, although it has not announced an exact release date or additional pricing details.
Featured image credits: Wikimedia Commons
For more stories like it, click the +Follow button at the top of this page to follow us.