Box Integrates Nvidia Nemotron to Power AI Agent Search

Box is bringing Nvidia’s Nemotron technology into its content management platform to help AI agents find information faster and deliver more accurate answers. The new feature, called Box Content Retrieval, was unveiled at Box’s developer conference on Thursday and is designed to improve how AI agents search through large collections of business documents.

By integrating Nvidia’s Nemotron embedding system, Box aims to make information retrieval more efficient while reducing the time and computing resources required to answer user queries.

According to Nvidia, the improvements could be significant. Kari Briski, Nvidia’s vice president of generative AI software, told TechCrunch that the technology is expected to deliver answers up to four times faster, reduce errors by 20% and cut token usage by 90%.

Box plans to make Content Retrieval available to all Enterprise Advance customers, with the rollout expected to begin early next year.

How Nvidia Nemotron Improves AI-Powered Search

Traditional search systems often depend heavily on matching keywords to find relevant information. However, this approach can struggle when users ask questions using different wording from the documents they are searching.

Box Content Retrieval takes a different approach by using metadata generated through Nvidia’s Nemotron embedding technology. These embeddings help AI systems understand the meaning and relationships between pieces of information rather than simply matching individual words.

This allows AI agents to identify relevant sections of documents without having to process entire files from beginning to end.

Box CEO Aaron Levie explained that the technology can improve the relevance of information returned to AI agents.

“Instead of an agent having to kind of read through each document, we can send back chunks from the documents that are most relevant to the user’s query or whatever the task is that the agent is doing,” Levie told TechCrunch.

In practical terms, this could help businesses reduce unnecessary document processing and get more useful answers from their internal information systems.

Nemotron Embeddings Bring More Context to AI Agents

The new feature is built around Nvidia’s Nemotron Embed project, which provides open-weight embedding models designed to improve automated information retrieval.

Embedding models play an important role in AI systems. They convert information into numerical representations that help machines understand relationships between words, documents and broader concepts. AI agents can then use this information to locate relevant content more effectively.

Although large AI companies have often focused more attention on other parts of the AI technology stack, embedding models are becoming increasingly important as businesses adopt AI agents for everyday tasks.

Nvidia has been expanding its open-source model efforts in this area. In July, its latest Nemotron release topped the RTEB benchmark, which measures the quality of information retrieval systems.

By bringing this technology into Box, the companies want to make it easier for AI agents to work with large amounts of business content.

Legal Teams Could Benefit From More Accurate Retrieval

Box believes the technology will be particularly useful for organizations that actively use AI agents to complete complex tasks.

One example is the legal industry. Lawyers frequently need to search through large collections of contracts, legal documents, clauses and previous cases to find information relevant to their work.

With Nemotron-powered embeddings, an AI agent could identify related legal concepts even when the documents use different words or expressions.

For example, a search for a particular legal concept could surface relevant clauses and precedents without requiring an exact keyword match. This could make document research more efficient and help legal professionals locate useful information more quickly.

Box Combines AI Models With Its Existing Infrastructure

Although Nvidia’s Nemotron models are freely available, recreating Box Content Retrieval independently may not be straightforward.

That’s because the system also depends on Box’s existing metadata infrastructure and permission controls. These features help ensure that users and AI agents only access information they are authorized to view.

This combination of technologies is an important part of Box’s approach. Rather than simply offering an embedding model, the company is integrating several capabilities into a single retrieval system.

Levie highlighted the complexity behind the feature, telling TechCrunch, “We’re probably packing in five different technologies into one simple API.”

With the growing adoption of AI agents across businesses, companies need reliable ways to connect these systems with their internal data. Box’s partnership with Nvidia reflects that shift, as both companies look to make enterprise information easier for AI agents to access and understand.

The planned rollout early next year will give Enterprise Advance customers access to the new retrieval capabilities, potentially making AI-powered document search faster, more accurate and less expensive.

Also read : TechCrunch Founder Summit 2026 Kicks Off Today in Boston at 9 AM ET