Overview
- Box announced a new feature called Box Content Retrieval that builds Nvidia’s Nemotron embedding models into its core product to help AI agents answer queries more efficiently.
- Nvidia and Box provided projected gains including faster query times, fewer errors, and big reductions in token use, but those performance figures are vendor projections and have not been independently verified.
- The implementation uses Nemotron to parse documents, create semantic embeddings, and rerank results so agents match concepts instead of relying on keyword search.
- Box plans to roll the feature out to its Enterprise Advance customers in early 2027 and says its existing metadata, indexing and permission controls will govern what agents can access.
- Because Nemotron is open-weight but Box combines it with its metadata, permissions and multi-technology API, the companies say the packaged capability will be hard for others to reproduce and is suited to domain workflows such as legal research.