Box and NVIDIA expand partnership on enterprise AI

The collaboration introduces Box Context Retrieval, an engine powered by NVIDIA Nemotron models.[2] Box says it can find permission-aware information across millions of files, including scanned documents, images, audio and video.[2] Its approach combines keyword matching with semantic search to identify passages relevant to an AI agent’s task.[3] Box also says faster indexing and lower latency can make new and updated content available to agents in real time.[2]

Box says customers could improve search accuracy by 20 percent, reduce token use by 90 percent and increase question-answering speed fourfold with Context Retrieval.[2] These are stated as potential improvements, not as guaranteed results for every deployment.[2] Box plans to make the retrieval engine generally available to Enterprise Advanced customers early next year.[2] Teams evaluating it will need to distinguish those projected outcomes from measured performance in their own content environments, and confirm timing and fit with Box as availability approaches.

NVIDIA Nemotron models, starting with Nemotron 3 Ultra, are available in Box AI to Enterprise Advanced customers.[2] The partnership is also intended to bring NVIDIA’s frontier-scale open-weight Nemotron models to Box.[2] Box says access to open models gives customers more choice, and its multi-model approach lets them select a model for each task based on performance, cost and control.[3][2] Separately, Box and NVIDIA are collaborating on agent security through NVIDIA OpenShell and Box Mount.[2] Box says OpenShell’s secure runtime, policy enforcement and supervision are intended to support safer agent workflows.[2]

Box Mount is planned to let agents work directly with Box folders inside isolated computing environments.[3] Box says agents will be able to read, edit and create files while respecting the source content’s permissions and security classifications.[3] The new features are set to be unveiled at BoxWorks 2026 in San Francisco.[2] At BoxWorks, NVIDIA CEO Jensen Huang discussed how Nemotron open models paired with business context and tools can help developers build agents that use enterprise knowledge.[4] Box also introduced Box Architect, a separate tool for describing processes in plain language to build automated workflows; its beta rollout to Enterprise Advanced customers was set for the fourth quarter.[1]