Introducing the new Box Agent

We are incredibly excited to announce the general availability of the Box Agent, enabling enterprises to transform their content into the exact context AI needs to reason, create multi-step plans, and complete complex work with simple, natural language direction.

The majority of enterprise data is unstructured, and harnessing actionable insights can mean hours of manual searching and analysis. The Box Agent addresses these challenges by shifting from simple keyword search to true, action-oriented intelligence that understands intent and delivers grounded answers. It combines multi-step reasoning with Secure RAG (retrieval-augmented generation) to ensure every outcome is grounded in your actual enterprise files, folders, and metadata. We do this with a combination of:

• The latest AI models, from Open AI, Anthropic, Google Deepmind, and more to power diverse reasoning tasks

• Multi-step reasoning and planning that allows the agent to create a plan, execute multiple steps, and iterate on its own to solve complex problems

• Content grounding with verifiable outputs that provide direct citations linking back to the source documents in Box for confident validation

• Permissions and governance, ensuring the Agent natively respects strict security permissions, retention rules, and compliance controls by default

   

Users can interact with these capabilities through AI Home, a new, full-screen conversational interface that serves as a central hub for all agent interactions. AI Home supports “work sessions” where users can browse conversation history and refine work in progress.

For Enterprise Advanced customers, Box AI Studio provides an administrative tool to build bespoke, task-specific agents tailored to unique enterprise workflows. Within AI Studio, process owners can configure:

• Model provider selection to choose the best underlying LLM for the task

• Custom prompts and instructions to outfit agents with specific knowledge or unique guidelines

• Access governance to select exactly which users or groups can access specific custom agents

Customers can choose between different agent configurations depending on their reasoning requirements:

• Standard Mode (Default): Optimized for everyday, simpler tasks and faster answers

• Pro Mode: Designed for complex tasks with a high degree of reasoning that require the agent to think longer for deeper answers

   

For high-scale needs, admins can leverage Expanded Mode in AI Studio to remove standard query length limits, allowing agents to run long-running queries against a large corpus of data. Usage in Expanded Mode is chargeable and draws from the organization’s pool of AI Units.

The Box Agent enables customers to implement high-value, content-driven use cases on a single secure and compliant platform. By transforming content from static storage to active intelligence, enterprises can:

• Accelerate decision-making by synthesizing insights across multiple files automatically.

• Reduce manual work by drafting summaries, reports, and proposals grounded in company content.

• Maintain strict security by keeping enterprise data safe within Box’s secure boundaries without training external vendor models.