Box unveiled Box Content Retrieval, which uses Nvidia Nemotron embeddings to help AI agents search company documents. The system will roll out to Box Enterprise Advance customers in early 2027.
Box presented Box Content Retrieval at its developer conference, showing how agents can receive relevant document extracts rather than read entire files. Nvidia’s VP of generative AI software, Kari Briski, expects answers to arrive 4 times faster, with 20% fewer errors and 90% fewer tokens used. These are estimates, not measured outcomes reported in the source.
The feature uses Nvidia Nemotron Embed, a project offering open-weight embedding models for automated retrieval. Its embeddings map underlying concepts beyond keywords, helping searches find related material even when wording or language differs.
For companies building AI agents, the distinction is between access to a model and a working document retrieval system. Although Nemotron models are freely available, Box also relies on its existing metadata and permission controls, making the complete system harder to recreate.
The performance figures are expectations, and the source provides no test results to validate them.
Companies using Box could give agents relevant document extracts without having to build the retrieval system themselves.
Freely available embedding models do not remove the need for metadata and permission controls.
Written automatically by the Made in 10 Newsroom from the source above.
Isomorphic Labs is in talks to raise about $5bn at a valuation of about $50bn, Bloomberg reported. The financing has not closed, and the AI drug discovery company's work remains preclinical.
Google says AI Mode contributed to weaker-than-expected search revenue at parent Alphabet. The AI search service monetises at a lower rate than traditional search.
Sarvam AI has launched Prism, an open-weight AI model for making structured decisions in business workflows. Its Apache 2.0 licence lets developers adapt the model and deploy it on their own infrastructure.