Box has announced Context Retrieval, an enterprise search layer using NVIDIA’s Nemotron models. It is designed to give AI agents relevant document passages while preserving access permissions and file metadata. Box plans to offer the engine to Enterprise Advanced customers early next year, with legal clauses expressed in different wording among its stated uses.
Box, the enterprise content management company, has announced Context Retrieval, an engine using NVIDIA’s Nemotron models to find relevant business content for AI agents. The design combines retrieval with Box’s existing permission controls. It is intended to give an agent information from content that the user may access, with a planned rollout to Enterprise Advanced customers early next year.[1], [2]
Instead of passing an entire file to an agent, Box aims to return the document passages relevant to its task. Existing metadata supplies additional information about the files. Aaron Levie described combining several technologies behind one application programming interface. The retrieval engine therefore includes Box’s surrounding infrastructure as well as NVIDIA’s embedding models, which represent concepts numerically.[1]
The company illustrates the design with legal work: an agent could find relevant contract clauses even when they use different wording. Embedding models are intended to match related meanings rather than only identical keywords. Box says the surrounding access controls continue to determine which content can be retrieved. The announcement identifies an enterprise customer group and a future delivery period for the new engine.[1]
