Agents in OneDrive

Agents in OneDrive is a feature that lets you create a persistent AI agent inside OneDrive that can answer questions, summarize, and extract insights across a specific set of files you choose. Instead of chatting about one document at a time, you “bundle” a project’s documents into an agent that keeps that context available over time. This is the practical difference between “ask about this file” and “ask about this project.” For teams trying to operationalize AI inside real document workflows, an AI certification helps because it forces clarity around scope, access control, and what the model is actually allowed to see.

What it is?

An agent in OneDrive is represented as a file with a .agent extension. It appears in your OneDrive file list like any other item, typically with a Copilot-style icon. You open that .agent file directly in OneDrive, and it launches a full-screen chat-style viewer. The agent’s responses are grounded in the specific documents you attached when you created it.

The key idea is persistence. The agent is not a temporary chat session that forgets your project context. It is an object in your storage that remains available, shareable, and updateable like other files.

Where it works

As of the general availability rollout, the feature is available worldwide for OneDrive on the web. The web experience is the primary surface for both creating and using agents. That matters because it sets expectations: this is first shipped as a web-first OneDrive capability, not a broadly surfaced desktop or consumer feature.

Who can use it

This feature is available only to work or school accounts with the appropriate Copilot license. If you do not have that license, the creation and use flow described here will not be available.

This is a common enterprise reality: the feature can exist, but entitlement determines whether you ever see it in your interface.

How you create an agent

Creation happens from the OneDrive web file list. You use the main create action, typically labeled something like “Create or upload” or “+ Create,” and select “Create an agent.” After you complete the creation flow, the result is a .agent file placed into your OneDrive, which becomes the entry point for chatting with that agent.

That design choice is deliberate. By making the agent a file, the system inherits familiar behaviors like sharing, moving folders, and file lifecycle management.

What content it can include

The agent is built from a curated set of files you select. It is meant to work across multiple documents in a single context rather than being tied to one file.

At general availability, multiple reports describe a limit of up to 20 documents per agent. In practice, this means you create one .agent file that represents a bundle of up to around 20 project files, giving you a manageable “project brain” rather than an unbounded search across your entire drive.

What you can do with it

The core value is cross-file reasoning without manual file switching. Typical uses include:

Asking questions across a project’s documents such as plans, meeting notes, specifications, and presentations.

Generating summaries of progress, major decisions, and current status from multiple sources.

Extracting operational details such as deadlines, action items, responsibilities, or open questions scattered across documents.

The point is to get insights across multiple files without opening each one, which is especially useful when information is distributed across a chaotic pile of docs like it always is in real organizations.

Sharing and collaboration

Agents are shareable like other files in OneDrive. If you share the .agent file, collaborators can use the same agent context, provided they have permission to access the underlying documents that were used to build the agent.

The permission boundary is the critical part. Sharing the agent does not bypass document access. If someone cannot access a source file, the agent should not become a loophole that exposes that content. Instead, the agent operates within the same permission model as the files it is built from.

This makes agents useful for teams because everyone can use the same project context while still respecting organizational access controls.

Updating an agent

Agents are designed to remain relevant as projects evolve. You can update an agent by changing the documents it references and by refining the agent’s instructions. This matters because most real projects change daily. New docs appear, priorities shift, and yesterday’s summary becomes today’s misinformation.

Treat the agent like a maintained project asset. If you never update its document set, it will slowly drift away from reality, just like every wiki page humans stop maintaining.

Admin and licensing gotchas

Even when an organization has the required Copilot license, service plan configuration can block the feature from appearing. Some rollout notes point out that a specific service plan associated with SharePoint must be enabled inside the Copilot license for the option to show up. There is also confusion in some environments between this capability and separate agent monetization or pay-as-you-go setups.

Translation: organizations can be “licensed” and still functionally disabled, because admin toggles exist specifically to keep IT departments employed.

Security and governance claims

The feature is positioned as operating directly over organizational files while respecting existing security boundaries. In practical terms, that means the agent should only work with content a user is allowed to access, and the organization’s controls, permissions, and data boundaries continue to apply.

What changes is not what you can access, but how quickly you can interpret and summarize what you already have access to.

What it is not

This is not a consumer OneDrive feature at launch. It is scoped to work or school accounts with the required Copilot licensing.

It is also not the same as generic “ask about this file” Copilot experiences. The distinctive value is a persistent agent tied to a chosen bundle of documents, represented as a .agent file that can be reused, shared, and updated over time.

Known gaps and what is still evolving

This is an early generally available release and it is expected to evolve. Areas likely to change include broader surface support beyond OneDrive web, deeper automation and agent behaviors, and more extensibility for teams.

The important takeaway is that the foundation is now in place: a file-based, shareable agent object that represents a maintained project context rather than a one-off chat.

Conclusion

Agents in OneDrive turns a collection of documents into a persistent, reusable AI assistant that lives alongside your files. The .agent file format makes it manageable like a normal OneDrive asset, while the curated document bundle makes it useful for real projects instead of one-document trivia.

The feature’s real impact is workflow compression: fewer tabs, less hunting, faster synthesis, and a shared team context that stays in one place. Used well, it reduces the “where did we write that down” problem that has been humanity’s core operating system bug for decades.