Box and NVIDIA Expand Enterprise AI Retrieval and Agents
Box expanded its NVIDIA partnership with Context Retrieval, Nemotron model access and an AI workflow builder. Here is what is available and what comes next.
- Box announced Context Retrieval, powered by NVIDIA Nemotron models, to help AI systems find information across enterprise content.
- Box Architect is a new workflow builder, while NVIDIA OpenShell adds a security layer for agent execution.
- Box's performance figures are company claims, and several features have future rollout dates rather than immediate general availability.
Box announced an expanded NVIDIA partnership on November 5, 2026, at 9:00 a.m. Eastern Time, which was 7:30 p.m. in India. The update introduces Box Context Retrieval, brings more NVIDIA Nemotron models into Box AI and adds Box Architect, a natural language workflow builder. It matters to enterprise teams building agents that need to find and act on company content without bypassing existing permissions.
News Summary
The announcement brings together three needs that often slow down enterprise AI projects: finding the right company information, building repeatable workflows and controlling what an agent can access. Box says Context Retrieval will help AI systems find useful passages across company files. NVIDIA's Nemotron models provide part of the retrieval and model layer, while OpenShell is intended to help secure agent execution.
Reuters also reported that Box introduced Box Architect, which lets teams describe a process in plain language and build an AI driven workflow. Box says several parts of the update will roll out in stages. The announcement does not mean that every customer can use every feature today.
What Happened
Box said it is expanding its collaboration with NVIDIA. The main product announcement is Box Context Retrieval, a new retrieval engine powered by NVIDIA Nemotron models. It is designed to help AI systems search across enterprise content, including documents and other media, and return information that is relevant to a user's question or task.
The partnership also brings NVIDIA's open weight Nemotron model family to Box AI. Box's release highlights Nemotron 3 Ultra for enterprise customers. The companies are also extending security and governance for agents through NVIDIA OpenShell, a runtime intended to provide a controlled environment for agent actions.
Reuters reported a second product, Box Architect. It lets a team describe a process in natural language and create an AI workflow around that process. Reuters said Box planned to begin a beta rollout to Enterprise Advanced customers in the fourth quarter of 2026.
The announcement also includes new threat protection aimed at bulk downloads, account takeovers and mass deletion. Box said those protections are expected to reach Enterprise Advanced and Shield customers early next year.
Why It Matters
An AI agent is only useful when it can work with the information that a business already owns. That information is spread across contracts, spreadsheets, presentations, images, scanned files, recordings and other documents. If an agent cannot find the right content, it may give a weak answer or miss an important detail.
Giving an agent access to everything is not a safe solution. Companies need the same access rules for agents that they use for employees. A person who cannot open a confidential file should not be able to ask an agent to retrieve it on their behalf.
Box Context Retrieval is designed around this problem. It aims to improve retrieval while keeping the permissions attached to the content. OpenShell addresses a different part of the same challenge: controlling how an agent runs and what tools or resources it can use.
Box Architect adds a workflow layer. Instead of asking users to write code for every process, the product is intended to turn a plain language description into an AI workflow. That could make it easier for business teams to describe a task, but the final workflow still needs review, access rules and testing before it handles important company work.
Key Details
Box said Context Retrieval uses NVIDIA Nemotron models to find information across enterprise content. The company claims it can improve retrieval accuracy by 20 percent, reduce token use by 90 percent and make question answering four times faster. These numbers come from Box. The announcement does not provide a full independent benchmark report or enough testing detail to treat the figures as a result that every customer will see.
Box said Context Retrieval is expected to become generally available to Enterprise Advanced customers early in 2027. The release also says NVIDIA Nemotron 3 Ultra is available to Enterprise Advanced customers now. Access to a model or retrieval feature can still depend on a customer's account and configuration.
Reuters reported that Box Architect is expected to begin a beta rollout to Enterprise Advanced customers in the fourth quarter of 2026. The tool is described as a workflow builder that accepts a plain language description of a process. The report also notes that new threat protection against bulk downloads, account takeovers and mass deletion is expected early next year for Enterprise Advanced and Shield customers.
These dates matter. A product announcement is not the same as a general release. Teams should confirm what is available in their account before building a production process around it.
Who Is Affected
The update is most relevant to Box enterprise customers, especially organisations that want to use AI agents across internal files. IT teams may care about model access, security controls and rollout dates. Information security teams may focus on agent permissions and threat protection. Developers and automation teams may be interested in retrieval quality and the workflow builder.
Content owners may also be affected. If an agent can read documents, the organisation needs to know which users can ask questions about those documents and whether the agent respects the same access rules. A well designed retrieval system should not turn a private file into a public answer.
Regular users should not assume that the update changes their everyday Box experience. The announcement focuses on enterprise products and plans. Availability and pricing need to be confirmed with Box.
What Happens Next
Box says Context Retrieval is expected to reach general availability for Enterprise Advanced customers early in 2027. The Nemotron model availability is described as current for Enterprise Advanced customers. Box Architect is expected to enter beta in the fourth quarter of 2026, according to Reuters.
Customers should ask their Box account team for the exact availability in their environment. They should also review the security model before connecting external agents or allowing automated changes to company content.
A sensible pilot should begin with read only tasks. For example, an agent can find relevant clauses in approved contracts and provide links to the source documents. A person can check the answer before any change is made. After that, the team can test workflows that create drafts or prepare updates, with human approval before a final action.
What We Know and Do Not Know
We know that Box announced an expanded partnership with NVIDIA on November 5, 2026. We know that Context Retrieval is built around Nemotron models, that Box Architect is a workflow builder and that OpenShell is part of the security direction. Box and Reuters both describe staged availability for the new capabilities.
We do not know whether the vendor performance claims will hold across every customer dataset. The announcement does not provide an independent evaluation covering all file types, languages and business use cases. It also does not publish a universal price for the new features or a single availability date for every customer.
The product roadmap can change. Enterprise teams should treat the dates as announced plans and confirm them with Box before making purchasing or deployment decisions.
Key Takeaways
- Box Context Retrieval is designed to help enterprise agents find relevant information while respecting permissions.
- Box Architect adds a plain language workflow builder, with a beta rollout expected in the fourth quarter of 2026.
- The 20 percent, 90 percent and four times performance figures are Box claims, not independent results.
Related Reading
FAQ
What did Box announce with NVIDIA on November 5, 2026?
Box announced Context Retrieval, wider Nemotron model access, security work using NVIDIA OpenShell and Box Architect, an AI workflow builder. The goal is to support enterprise agents that need access to company content.
What is Box Context Retrieval?
It is a retrieval engine powered by NVIDIA Nemotron models. It is designed to help AI systems find useful information across enterprise files while respecting the access permissions that apply to those files.
Is Box Architect available now?
Reuters reported that Box Architect is expected to begin rolling out in beta to Enterprise Advanced customers in the fourth quarter of 2026. Confirm the current status with Box before planning a production workflow.
When will Box Context Retrieval be generally available?
Box said general availability for Enterprise Advanced customers is expected early in 2027. The date is a company plan and may change.
What performance improvements did Box claim?
Box claimed 20 percent better retrieval accuracy, 90 percent lower token use and four times faster question answering. Those figures should be treated as vendor claims because the announcement does not include an independent benchmark report.
Does this update affect regular Box users?
The announcement focuses on enterprise customers and agent workflows. It does not describe a new general consumer feature available to every Box account.