Search accuracy could improve by 20%... Box unveils enterprise AI engine

Box unveiled Box Context Retrieval, an enterprise AI search engine built on NVIDIA's Nemotron models, claiming it can improve search accuracy by 20%, cut token usage by 90%, and speed up query responses fourfold. The company announced the engine and an expanded NVIDIA partnership at BoxWorks 2026, with CEO Aaron Levie framing secure access to company-specific context as the key to unlocking enterprise AI. The engine locates relevant passages in corporate documents and feeds them to AI agents, targeting the retrieval layer that determines whether agentic AI can be trusted on proprietary data. The claims matter because token efficiency and retrieval precision are central cost and reliability constraints for enterprise AI deployments, and the tie-up deepens NVIDIA's push into enterprise software. What to watch next is independent validation of the 20% accuracy and 90% token-reduction figures, plus pricing and general availability for Box's enterprise customers.