The Pulse
Box Announces NVIDIA-Powered Context Retrieval Engine for Enterprise AI Agents
Box announced Box Context Retrieval, an engine powered by NVIDIA Nemotron models, as part of an expanded collaboration with NVIDIA. Box says customers will have the potential to improve search accuracy by 20 percent (Recall@5), reduce token
AI.info Team · Written with AI · fact-checked ·
Box announced an expanded collaboration with NVIDIA that it says enables organizations to securely execute AI-powered workflows at scale. Box is introducing Box Context Retrieval, a new engine powered by NVIDIA Nemotron models, and is bringing NVIDIA’s Nemotron family of open-weight models to Box AI.
Aaron Levie, CEO and co-founder of Box, said: “By partnering with NVIDIA, we are creating a new intelligence layer for enterprise content that gives customers faster, more precise ways to ground AI in the content that powers their business. Open weight AI is an important part of that vision, complementing closed models while driving innovation and broader adoption across the industry. Together, these capabilities give enterprises more choice in how they build AI workflows while preserving the security that Box is known for.”
The engine will retrieve across millions of files
Box Context Retrieval uses NVIDIA Nemotron Retriever, a collection of open models that Box says turns enterprise data into a structured knowledge layer for agents to reason over. With Nemotron Retriever embedding models, the engine will enable agents to retrieve relevant, permission-aware information across millions of files. That includes content that has historically been difficult for AI to interpret, such as scanned documents, images, audio and video.
Customers will have the potential to improve search accuracy by 20 percent (Recall@5), reduce token usage by 90 percent and accelerate question-answering speeds fourfold, Box says. The company describes retrieval as permission-aware: an agent receives information only from content the user is permitted to access. Significantly faster indexing and lower latency will make new and updated content available to agents in real time, Box also says.
Hybrid search and open-weight models
The retrieval engine includes hybrid search, which Box says combines precise keyword matching with an understanding of meaning and context, so users can find material without knowing the exact words to search. Box gives examples of professionals using the system to find contract clauses across files with different terminology, or to analyze insurance claim packages containing scanned forms, property images, recorded statements and video. The company also describes uses in financial analysis, clinical research and university administration.
The collaboration will also bring NVIDIA’s frontier-scale open-weight Nemotron models to Box, including NVIDIA Nemotron 3 Ultra. The multi-model approach lets customers select a model for each task based on performance, cost and control. Box presents the open models as an option for specialized, high-volume workloads, with content continuing to be managed in Box.
Box and NVIDIA collaborate on securing agent execution
Box and NVIDIA are also extending security and governance to agent execution through NVIDIA OpenShell. The companies are members of the Open Secure AI Alliance, and the collaboration brings together Box’s content security and governance capabilities through Box Mount with OpenShell’s secure runtime, policy enforcement and supervision. The work is intended, Box says, to help enterprises deploy agents with greater control over how they access content, use tools and execute tasks.
Box Context Retrieval will be generally available to Enterprise Advanced customers early next year. NVIDIA’s Nemotron models, beginning with 3 Ultra, are now available in Box AI to Enterprise Advanced customers.