Box AI Agents

Box AI Agents use top foundation models like GPT-4 or Claude, selected and orchestrated by Box, to understand, reason, and generate. Each model is matched to the use case to ensure performance, accuracy, and enterprise alignment.

Every Agent is designed around a clear objective, like summarizing a document or identifying legal risk. This goal defines what the Agent should accomplish, keeping its output targeted, actionable, and relevant to your business.

Instructions act as the agent’s strategy, structured prompts or step-by-step logic that guide how it achieves its goal. These can define output format, tone, or workflows for consistency across departments or roles.

Agents apply advanced skills like summarization, Q&A, and classification, and can call Box-native or external APIs. This allows them to not just understand content, but take action and generate outcomes in the flow of work.

Box AI Agents are grounded in your real enterprise content—files, folders, and metadata—using secure, permission-aware RAG. They only access what users are allowed to see, and always stay within Box’s governance boundaries.

Customize agents with the latest and most powerful AI models from providers like OpenAI, Anthropic, and Google. Box AI gives you the flexibility to choose the right model for each task, so you can future-proof your AI strategy without getting locked into a single vendor.

Give agents the context they need to perform. Attach relevant documents or Box Hubs to enhance understanding, accuracy, and context. Customize behavior with custom instructions tailored to your business.

Experiment and refine with custom instructions. Mix and match models, tweak instructions, and fine-tune your agents to discover the best combination for your enterprise’s unique use cases. 

We know it’s important for admins to maintain granular access controls, so we extend our AI Principles to Box AI Agents. With Box AI Studio, you can restrict access to only the users who need it and ensure secure AI deployment.