The Box Agent uses state-of-the-art foundational models from OpenAI, Anthropic, Google, and more to understand, reason, and generate. 

The Box Agent is designed for executing on a clear objective, like summarizing a document or identifying legal risk. That way you can ensure you’re accomplishing your goal through targeted output that’s actionable, and relevant to your business.

The Box Agent uses natural language to reason, create multi-step plans, and complete complex work. By leveraging implicit multi-agent orchestration, a single prompt strings together finding files, analyzing data, and creating new deliverables.

Apply advanced skills like summarization, Q&A, and content drafting directly within your workflow. The Agent chains together native tools to transform unstructured data into final outcomes, like executive summaries or presentations.

Built natively on the Box content layer, the Agent operates with permissions-aware retrieval to ensure outputs are based only on content a user is authorized to access. It inherently maintains all existing security permissions, retention rules, and compliance controls on your content.

Empower your business with LLM choice by selecting the best foundational model—such as Gemini, ChatGPT, or Claude—for every task. Box AI provides the flexibility to future-proof your strategy and optimize performance across diverse use cases without vendor lock-in

Ground AI in the content that runs your business. Using Box AI Studio, administrators can outfit custom agents with unique instructions and scoped knowledge, like specific files or Box Hubs, to ensure every output is contextually accurate and tailored to your specific workflows.

Experiment and refine with precision in Box AI Studio. Tweak instructions, swap models, and test your custom agents against real enterprise context to discover the optimal configuration for your unique business needs before deploying at scale.

The Box Agent is secure by design, maintaining existing user permissions, retention rules, and compliance controls on your content by default. Admins maintain granular authority through Box AI Studio, ensuring that agents only access authorized content and that data is never used to train external models.