OneDrive Agents: AI-Powered Project Context with Microsoft 365 Copilot

Microsoft has launched Agents in OneDrive, a new feature integrated with Microsoft 365 Copilot that allows users to create AI-powered assistants tailored to specific projects and sets of documents. The feature, generally available worldwide as of today, February 4, 2026, aims to streamline workflows and improve information access by enabling users to ask questions and receive summaries grounded in their own content.

Rather than repeatedly querying Copilot across individual files, users can now build an “agent” that understands an entire collection of documents – including project plans, specifications, meeting notes, research, and presentations. Microsoft describes these agents as “focused AI teammates” built from a user’s own files and folders.

The core functionality of an agent is to provide context-aware responses. According to Microsoft, agents can answer questions across multiple documents simultaneously, summarize key discussions and decisions, identify owners, deadlines, risks, and action items, and maintain a consistent understanding of evolving projects. This addresses a common pain point for teams working with large volumes of documentation, where finding specific information or recalling past decisions can be time-consuming.

Agents are saved as .agent files within OneDrive, functioning similarly to standard document or spreadsheet files. Opening an agent file launches a full-screen Copilot experience dedicated to that specific project or topic. Users can locate agents through OneDrive search, filtering by the “Agent” file type.

Getting started with Agents in OneDrive requires a Microsoft 365 Copilot license and access to OneDrive on the web. The creation process is designed to be straightforward, requiring no special administrative setup. Users can create an agent by selecting “+ Create” or uploading files and choosing “Create an agent” from the toolbar or right-click menu. Up to 20 files can be selected to form the knowledge base for a single agent. Users can also name the agent and add optional instructions to guide its behavior.

The technology leverages Microsoft Dataverse to index content while respecting existing file permissions, ensuring users only access information they are authorized to view. This addresses data security and access control concerns inherent in AI-powered document analysis.

Microsoft highlights several use cases for Agents in OneDrive. Teams can use them to quickly gain clarity on project status (“What decisions have we made so far?” or “What’s still open, and who owns it?”), onboard new team members by providing a readily available knowledge base (“Explain how this team operates based on these docs.” or “Summarize how we ship new features.”), and maintain momentum by quickly recalling past discussions (“What did we agree to in recent reviews?” or “What risks keep coming up?”). Agents can also be used to identify themes across reports or extract learnings from research materials (“What themes show up across these reports?” or “What did we learn from this research?”).

Unlike traditional file searches, Agents preserve context and history within a single interface. As projects evolve, users can update agents by adding or removing files or refining the initial instructions, ensuring the AI assistant remains aligned with the latest information. Sharing an agent with colleagues is as simple as sharing any other OneDrive file, provided collaborators have the necessary permissions to access the underlying source documents.

The launch of Agents in OneDrive builds on earlier work in the Copilot Studio, where OneDrive files and folders could be added as configurable knowledge sources beginning with a public preview in May 2025. This integration allows agents to index and utilize documents for more contextually relevant responses.

Microsoft is actively soliciting feedback on Agents in OneDrive through standard Microsoft feedback channels, encouraging users to share their experiences and suggestions for improvement. The company specifically asks users to share what types of work they are using agents for, what insights they have gained, and what features would make the tool even more valuable.