Box Context Retrieval is a new enterprise AI search engine powered by Nvidia’s Nemotron models. Announced November 5, 2026, the technology helps AI agents locate information across company documents without processing entire files.
Box claims the system can deliver answers four times faster, improve retrieval accuracy by 20% and reduce token usage by 90%. These performance claims have not been independently verified across customer environments.
Box Context Retrieval Uses Nvidia Nemotron Models
The new engine integrates Nvidia Nemotron Retriever, a family of open-weight AI models. Its embedding technology converts document content into numerical representations, enabling searches based on meaning rather than exact keywords.
AI agents can retrieve relevant passages instead of reading entire documents. Box combines this capability with existing document metadata, security controls and access permissions.
The technology supports searches across contracts, scanned documents, images, audio and video. Legal teams, for example, could use it to locate contract clauses and related information.
Box CEO Aaron Levie said the system combines several retrieval technologies through one interface. This approach reduces the integration work developers need when building enterprise AI applications.
Faster Search Could Reduce AI Processing Costs
AI agents consume computing resources when processing large volumes of information. Retrieving only relevant passages can reduce token usage, response times and operating costs.
Nvidia has also published technical benchmark results for Nemotron 3 Embed, describing improvements in retrieval accuracy and agent efficiency. Its 8-billion-parameter model ranked first on the RTEB leaderboard in Nvidia’s July 2026 evaluation.
Nvidia’s broader enterprise AI efforts include reasoning models developed with Salesforce for business applications.
However, Box’s performance estimates depend on document complexity, search requirements and implementation. Enterprise customers will need to evaluate the results against their existing systems.
Box Plans Enterprise Rollout in Early 2027
Box Context Retrieval is scheduled for general availability to Enterprise Advanced customers in early 2027. The company has not announced separate pricing for the feature.
Box is also expanding access to Nvidia’s open-weight Nemotron models through Box AI. Its partnership includes AI agent security controls through Nvidia OpenShell and Box’s existing governance infrastructure.
The integration builds on Box’s enterprise content management platform. TechAmerica previously reported on Anthropic’s workplace integrations, which connect Claude with Box and other business applications.
The upcoming rollout will help determine whether Box delivers its claimed improvements in real-world enterprise environments. Search accuracy, processing costs and permission controls will remain important measures of performance.
