AI
Box embeds Nvidia Nemotron into its system for agentic content retrieval
Box released Box Content Retrieval, integrating Nvidia’s Nemotron embedding model into the core of its product and using metadata to help AI agents retrieve enterprise content faster, more accurately, and at lower cost. Box said the feature is aimed at Enterprise Advance customers and is planned for early next year.
Box showcased Box Content Retrieval at its developer conference. The feature integrates Nvidia’s Nemotron embedding system into the core of Box’s product, using embedding metadata generated by Nemotron to help AI agents retrieve content more effectively when answering questions or performing tasks.
Box said users will get faster, more accurate, and lower-cost queries. Kari Briski, Nvidia’s vice president of generative AI software, told TechCrunch that query speed is expected to improve by up to four times, errors to decrease by 20%, and token usage to fall by 90%. The feature will begin rolling out to Box Enterprise Advance customers early next year.
Box CEO Aaron Levie said the system can return the document snippets most relevant to a user’s query or an agent’s task without requiring the agent to read documents one by one. It is powered by the Nvidia Nemotron Embed open-weight embedding model; the recently released Nvidia Nemotron previously led the RTEB benchmark, which measures retrieval quality.
Box said the system is especially well suited for use with active agents. For example, lawyers can more easily find relevant clauses and precedents because Nemotron embeddings do not merely match keywords but can map underlying concepts, discovering related legal topics even when wording or languages differ. Although the Nemotron model is freely usable, the solution also relies on Box’s existing metadata and permission controls, which Levie said amounts to packaging multiple technologies into a simple API.