At Build 2025, we introduced the Microsoft 365 Copilot APIs. Developers now have an extensibility layer that empowers them to create intelligent, AI-driven experiences deeply integrated with the Microsoft 365 ecosystem. These APIs unlock powerful capabilities like content retrieval, meeting summarization, and user interaction export and accelerate innovation while ensuring enterprise-grade security and compliance. 

Today, we’re excited to announce that the Retrieval API, one of the most requested APIs in the set, is now available in Public Preview.  

The Retrieval API gives developers a secure, compliant and scalable way to integrate enterprise content into their AI workflows. Whether you’re building RAG pipelines, domain-specific agents or mining insights, it provides a powerful foundation for grounding large language models in trusted organization’s knowledge with minimal setup and maximum flexibility.  

Early adopters such as Accenture, Thomson Reuters, Peppermint, Templafy, and Petrobras are already putting the Retrieval API to work and powering value across a range of enterprise AI use cases. 

“At Accenture, we view the Retrieval API as a transformative advancement in our multi-platform agentic strategy. It extends the capabilities of our substantial Microsoft 365 Copilot footprint by enabling intelligent, context-aware agents to operate seamlessly across various environments. This API allows our custom applications and agents to securely access Microsoft 365 data while maintaining existing permissions, simplifying development and ensuring built-in compliance.” – Accenture  

Why it matters: Built for enterprise-grade AI

Building AI experiences in the enterprise isn’t just about model quality; it’s about trust, scale, and seamless integration with how your organization works. The Retrieval API is designed to meet that bar, enabling developers to innovate confidently with secure, scalable access to enterprise knowledge.  

The Retrieval API enables natural language queries over Microsoft 365 content, returning relevant, permission-trimmed snippets ready to ground AI experiences. With a simple REST call, developers can retrieve knowledge from SharePoint Online and Copilot connector–indexed content, making it easy to build intelligent, context-aware solutions. 

“As a customer who is heavily invested in Microsoft’s storage, SharePoint Online and OneDrive, the main benefit of the Retrieval API is relieving us of the weight of managing content indexing. With the Retrieval API, we can maintain our own post processing flows and create agents while still being able to access raw content and leverage it through the gen AI platform of our choice.” – Petrobras  

Here’s what makes the Retrieval API stand out:  

• Secure and trustworthy by design: All results are permission-trimmed at query time, reflect Microsoft Entra ID access controls, and carry sensitivity label metadata, ensuring your agents and orchestrators never surface unauthorized or misclassified content.  

• Less infrastructure overhead: You no longer need to build and maintain your own retrieval infrastructure or vector indexes. The Retrieval API uses the same semantic index that powers Microsoft 365 Copilot, so you can ground LLM responses without additional pipelines or duplication.  

• Developer control with enterprise flexibility: With support for custom filters using metadata fields and connector labels, the API gives developers fine-grained control over what content is retrieved and how it’s used in downstream logic.   

• Faster performance with fewer API calls: The Retrieval API supports JSON batching, allowing multiple queries to be sent through a single request. This makes it easier to power concurrent grounding workflows while keeping API calls efficient and scalable in large applications.  

The Retrieval API brings the semantic richness of Microsoft 365 to your AI stack, without compromising security, governance, or developer agility.  

What can you build with the Retrieval API

The Retrieval API unlocks powerful, real-world scenarios by providing a secure, scalable semantic access layer to enterprise content. It expands what’s possible for pro-code developers, enabling use cases that previously were only possible with low-code tools like Copilot Studio through Tenant Graph grounding, while still supporting a range of approaches to building agents in Microsoft 365. 

“Thomson Reuters is pioneering the integration of its AI assistant, CoCounsel, with Microsoft 365 via the Retrieval API to enhance legal workflows. In document review scenarios, CoCounsel enables legal professionals to query batches of contracts stored in SharePoint or OneDrive—extracting key clauses like indemnity or confidentiality—grounded in actual document content for accuracy and compliance. Simultaneously, CoCounsel can search and pull value from internal legal databases and Microsoft 365 content to answer complex legal queries, such as identifying standard clauses or retrieving documents tied to specific clients or legal issues. Thomson Reuters early adoption underscores its commitment to transforming legal operations through AI, without compromising on security, context sensitivity, or compliance.” – Thomson Reuters  

Here are a few examples of what’s possible with the Retrieval API:  

• Policy and process agents: Build agents that can answer questions like “What’s our latest expense policy?” by retrieving relevant snippets from SharePoint and connector-indexed documents, ensuring responses are grounded in real, up-to-date internal content.  

• Meeting-aware workflows: Use the Retrieval API alongside the Meeting Insights API to surface documents or decisions related to a meeting topic, helping teams follow up more effectively without having to search manually.  

• Domain-specific knowledge assistants: Create agents tailored for legal, HR, or finance teams that pull content from curated sites and filtered fields using metadata, content source, or connector tags.  

• RAG-based applications: Use the Retrieval API as the retrieval layer in retrieval-augmented generation (RAG) pipelines, grounding model responses in tenant-specific content while preserving full compliance with organizational security and access policies.  

By combining natural language queries with live, governed enterprise data, the Retrieval API brings intelligence closer to where work gets done, without adding infrastructure burden or governance risk.  

Get started

The Retrieval API is now available for public preview via Microsoft Graph. Getting started is simple.  

Prerequisites:  

• You’ll need a Microsoft 365 tenant with Copilot licenses assigned to any users making API calls.  

• Make sure your tenant has access to SharePoint Online or indexed entities via Copilot connectors. 

•  The Copilot Retrieval API will adhere to Restricted SharePoint Search and Copilot connector staged rollout settings. 

Try it out:  

• Use Graph Explorer or any REST client to test requests against the `/beta/copilot/retrieval` endpoint.  

• Start with a simple POST request containing a natural language query and the data source you are targeting.  

• Access the API via client libraries included in the Microsoft 365 Agents SDK under the Microsoft.Agents.M365Copilot namespace. Libraries are available for .NET, TypeScript, and Python.  

• Explore filters, batching, and response fields to customize for your scenario.  

Provide feedback:  

• Microsoft Q&A is the site for technical questions and answers for Microsoft products. Use the Microsoft 365 Copilot Development tag to ask questions, connect with other developers, and submit answers. 

Learn more:  

Check out the Retrieval API documentation to learn about the capabilities available today for public preview.  

We’re excited to see what you build with the Microsoft 365 Copilot APIs, and we look forward to learning from your feedback.