Today, we're announcing the Box Agent — a new kind of AI built to turn your organization's content into action.
Available in the new Box AI Home, the Box Agent represents a fundamental shift in how content provides AI with the correct context. It’s a new way to use AI to interact with your organization’s content — not just by searching for it, but by putting it to work.
Context is the missing link
For the past two years, AI has captivated the world with its ability to write, code, and reason. But for enterprises, a critical gap has remained: AI models know the internet, but they don't know your business. They don't have access to your Q3 strategy, your indemnification clauses, or your brand's unique voice. 90% of your organization's data is unstructured content — the specs, contracts, financial statements, and reports that drive real decisions. In most enterprises, this data sits fragmented across silos, inaccessible to AI in any meaningful way. That context AI needs, lives in your files.
In fact, content is the essential resource models and agents need to deliver real, meaningful results from AI. While most popular AI tools act primarily as assistants, we are seeing the rise of independent agents that reason, create multi-step plans, and complete complex work. These agents use content as their primary context and produce files as their output.
The Box Agent unites the reasoning power of large language models with the proprietary truth of your content, in a secure platform that connects business context to AI execution — end to end.
A new way to work: Box AI Home
The Box Agent lives inside the new Box AI Home — a full-screen, conversational experience designed for the kind of work that actually matters. Start with a simple, natural language prompt. The Box Agent does the rest.
It doesn't just search for files. It orchestrates across them — finding the right content, extracting what's relevant, reasoning through complexity, and delivering finished outputs. It replaces manual, multi-document workflows with a single, intelligent flow of search, analysis, and creation. And because sessions are persistent, you can iterate, refine, and return to work-in-progress exactly where you left off.
No complex pipelines. No prompt engineering. No switching tools.
Built for every job
By default, the Box Agent employs an agentic loop that autonomously reasons through your request:
- Upfront Planning: The Box Agent breaks complex goals into manageable steps, creating a visible "to-do list" of its reasoning process.
- Capability Selection: It chooses the right tools — whether searching your entire body of content, extracting data points, or invoking a Custom Agent built in Box AI Studio.
- Agent-to-Human Collaboration: You can add and preview sources within the flow of work, ensuring the Box Agent has the right content without switching tabs.
- Autonomous Execution: The Box Agent synthesizes information into a final deliverable. Soon, with file generation capabilities, it will be able to create and save formatted Excel, Word, or PowerPoint files directly back to Box.
While baseline orchestration is incredibly powerful, enterprise work varies wildly. To handle the variety of what organizations need from AI, the Box Agent scales across two key dimensions: the complexity of the reasoning required, and the volume of AI Units required to successfully execute a task.
Standard Mode: Standard Mode handles the tasks that anchor your day. It is immediate, conversational, and always on for finding information, summarizing a document, or synthesizing key data points.
Pro Mode: Available to Enterprise Advanced customers, Pro Mode is built for work that demands more. It unlocks higher thinking levels by using premium models and optimizes the Box Agent for high-quality planning, execution, and refinement.
Expanded Mode: Some tasks aren't just complex — they’re long. Also available to Enterprise Advanced customers, Expanded Mode is for work that requires the longest query capacity — such as tedious, high-volume repetition or mission-critical, long-running processes — Expanded Mode increases the context window of the Box Agent to 4M tokens.
Pro Mode and Expanded Mode are fully independent. You can use either mode individually, or combine them for your most demanding tasks, enabling organizations to deploy the Box Agent across a wide variety of Enterprise needs.
The Box Agent in action
Here's how teams are already putting the Box Agent to work:
- Automate complex RFP responses: Sales Engineers upload 50-page RFPs and ask the Box Agent to draft responses. It autonomously locates the latest product documentation, messaging, and compliance guides, then produces high-quality drafts — without anyone manually hunting down source material.
- Review contracts against company policy: In-house lawyers ask the Box Agent to review a vendor MSA against standard terms. It identifies deviations from the company's standard playbook, serving as a rigorous first pass for the legal team.
- Generate personalized onboarding schedules: HR Managers generate tailored 30-day plans for new hires. The Box Agent pulls role-specific handbooks and roadmaps, then synthesizes them into a structured, ready-to-use schedule.
- Draft on-brand marketing assets: Marketers draft launch blogs by pointing the Box Agent to PRDs and brand guidelines. The output reflects the company's voice — without manually tagging sources as context.
- Synthesize customer feedback: Product Managers ask the Box Agent for the top feature requests from the past month. It scans support transcripts and feedback documents, then surfaces patterns — complete with citations.
These aren't isolated tasks. They're complete workflows, handled start to finish.
Operationalize expertise with Box AI Studio
For repeatable, high-stakes workflows, Box AI Studio lets anyone build custom agents — no code required. Administrators can create specialized agents tailored to specific business rules and knowledge sets, choose which leading foundation model powers them (including Gemini, ChatGPT, or Claude), and deploy them across the organization.
Configure a "Contract Reviewer" or "Brand Steward" once. Make that expertise available to every team member, every time. Enterprise Advanced customers can go further — removing usage restrictions and running long queries across large corpora of data for the most demanding enterprise workflows.
AI Studio turns institutional knowledge into a repeatable, scalable asset.
Deliver security and governance
Built on a foundation of enterprise-grade security, Box Agent works with your content, maintaining your existing permissions, retention, and compliance controls — so governance is never an afterthought. There's no need to export sensitive data into external AI systems, and your content is never used to train AI models.
Every response includes direct citations linking back to the source document in Box, so users can inspect, validate, and confidently act on AI-generated outputs — delivering high-quality outcomes that can be traced back to the underlying content.
The content layer for the agentic future
Box acts as the centralized content layer across your entire AI stack — a file system built for the age of agents. Because files are the native unit of work for agents, serving as both context and output, Box provides a single source of truth that seamlessly connects your content to people, agents, and applications.
The Box Agent serves as the core entry point to that vision: a seamlessly governed tool that orchestrates complex work directly on your content. Because it operates on your source of truth, every generated output is immediately wrapped in your existing permissions, retention policies, and compliance controls. And its reach is expanding via the Box MCP server. The Box MCP server is available today with new use cases growing day by day. Soon, the Box Agent will be callable from external platforms — meaning tools like Microsoft 365 Copilot, OpenAI, Slack, and Salesforce can securely invoke your enterprise content and write outputs back into Box. Consider a team using Microsoft 365 Copilot to analyze customer discovery notes and draft a business proposal. The Copilot Agent calls upon the Box Agent via MCP, and the finished draft is written back into Box — immediately governed by the same controls that protect everything else. Your teams keep the flexibility to work in their preferred tools. Your content stays where it belongs.
Box Extract strengthens this further. It pulls key intelligence from unstructured files — contracts, invoices, onboarding documents — and saves that knowledge as structured metadata attached directly to the source file in Box. Instead of re-reading every document from scratch, the Box Agent and third-party agents alike draw on that durable context to drive faster, more accurate workflows at scale. Because that metadata lives in Box, it inherits the same enterprise-grade security that governs everything else.
Regardless of which AI tools your teams use, Box keeps your content managed, secure, and connected — in one place.
Get started today
The Box Agent and Box AI Home are available now for Enterprise Plus and Enterprise Advanced customers, included at no additional cost within usage limits. Box AI Studio is available for Enterprise Advanced customers.
We're excited to learn how early adopters put the new Box Agent to work and are collaborating with customers to rapidly iterate and roll out new capabilities. You can learn more here on how you can get started using the Box Agent.
Box AI Agents FAQs
What are AI agents, and why do enterprises need them?
AI agents are software systems that can reason, plan, and complete multi-step tasks on your behalf — going far beyond simple question-and-answer.
For enterprises, the challenge is that general-purpose AI models are trained on public data and know nothing about how your business actually runs. Your real intelligence lives in your contracts, policies, financial reports, and the thousands of documents your teams create every day. Enterprise AI agents close that gap by grounding AI in your specific content, so you get answers and outputs that are accurate, relevant, and traceable — not generic.
What is Box Agent?
Box Agent is a secure, content-centric AI agent built directly into Box. You give it a natural language prompt — "summarize all HR policy updates from 2024" or "compare these three contracts by renewal terms" — and it handles every step: finding the right files, analyzing the data, and generating a final deliverable like a summary, brief, or presentation.
You don't configure pipelines or choose tools. Box Agent reasons, plans, and executes the work for you, turning your content from static storage into active intelligence.
How do AI agents work with documents and enterprise content?
Box Agent works directly on the content already stored in Box. When you submit a prompt, it automatically applies the right capabilities — search, Q&A, and content drafting — in sequence, without you needing to switch between tools. It finds the most relevant files you're authorized to access, synthesizes insights across them, and produces a grounded output with citations linking back to the source documents. Every answer is traceable, so you always know exactly where the information came from.
How does Box Agent handle complex, multi-step work?
You type one natural language prompt — Box Agent handles everything from there. It breaks your request into a plan, coordinates the right capabilities behind the scenes — search, analysis, drafting — and delivers a complete output. Sessions are persistent, so you can iterate, refine, and return to work in progress without starting over. For Enterprise Advanced users, Pro Mode enables a more capable AI model for even more demanding tasks, while Expanded Mode removes token restrictions for longer-running, content-heavy work.
Are AI agents secure for enterprise use?
Box Agent is secure by design, not by configuration. Because it's built natively on Box, it inherits your existing permissions architecture automatically — every user's access controls, retention rules, and compliance policies apply to AI by default. AI-generated answers are based only on content the user is authorized to see.
Your data never leaves Box, is never used to train external LLMs, and every Agent conversation is subject to the same retention policies and legal holds as standard Box content. For IT and compliance teams, that means no new attack surface and no shadow IT risk.
What compliance standards does Box AI meet?
Box AI builds in audit trails, inherited permissions, and prompt injection protection — and operates within Box's enterprise-grade compliance framework, which includes FedRAMP High, HIPAA, and SOC certifications. AI activity isn't a new compliance category to manage — it's automatically covered by the governance framework you already have in place.
For regulated industries like healthcare, financial services, life sciences, and the public sector, that means you can deploy AI confidently without introducing new risk.
Can I choose which AI model powers Box Agent?
Yes. Box Agent supports LLM choice — you can select Gemini, ChatGPT (OpenAI), or Claude (Anthropic) as the underlying model, and switch as the AI landscape evolves. There's no vendor lock-in and no infrastructure rebuild required. This flexibility means your AI strategy isn't dependent on a single model provider, and you can always use the best model for your specific use case.
How do I connect external AI tools to my Box content?
The Box MCP Server is a standardized, secure bridge that connects third-party AI agents — including Anthropic Claude, GitHub Copilot, Mistral Le Chat, Microsoft Copilot Studio, and Salesforce Agentforce — directly to your Box content on demand, so agents always work from the latest version of your files.
External platforms can query and use your Box data without raw file content ever leaving Box's governed environment. Your existing security policies and access controls are enforced automatically via OAuth authorization. This eliminates the cost and complexity of building custom one-off integrations for every AI tool your organization wants to use.
How is Box Agent different from other enterprise AI tools?
Most enterprise AI tools are app-centric — they apply AI to the tool you're working in, not to your content itself. Box Agent is content-centric: unlike app-centric copilots, it operates across every file, format, and workflow in Box, applying AI to enterprise content itself — not just the tools used to create it.
Box AI is model-agnostic (no vendor lock-in), natively permissions-aware (no new governance layer to build), and automatically chains search, Q&A, and drafting together from a single prompt — no manual configuration required. And because intelligence comes to your content rather than requiring content to move, employees have no reason to export sensitive data into unsanctioned tools.
How do I get started with Box AI Agents?
Box Agent is available to Box customers on Business plans and above. You access it directly from the left-hand navigation in Box, where a full-screen conversational workspace — AI Home — lets you start with a natural language prompt — no setup, no pipelines, no AI expertise required.
Enterprise Advanced customers also get access to Box AI Studio to build custom agents, along with Pro Mode and Expanded Mode for more demanding workloads. If you want to connect external AI platforms to your Box content, the Box MCP Server is available for your IT and developer teams to configure. The fastest way to see value is to start with a focused pilot: pick a document-heavy workflow your team does manually today, and let Box Agent handle it.
