Copilot Studio Knowledge Sources: SharePoint, Dataverse, and Custom APIs

One of the most powerful features of Copilot Studio is the ability to ground your agent's responses in your organization's actual data. Here's how the three main knowledge source types compare.

SharePoint Knowledge Sources

Best for: Unstructured content - documents, wikis, policies, FAQs

When you connect a SharePoint site, Copilot Studio indexes the content of pages and documents. The agent can then answer questions by retrieving relevant passages.

Setup: In your agent's knowledge settings, add a SharePoint URL. The agent will crawl and index content within 24 hours.

Limitations: Large SharePoint sites (10,000+ pages) may have indexing delays. PDF content is extracted but formatting isn't preserved. Password-protected documents are skipped.

Dataverse Knowledge Sources

Best for: Structured data - records, tables, relationships

Connect your agent to Dataverse tables and it can answer questions like "How many open cases do we have?" or "What's the status of order #12345?"

Setup: Select the Dataverse tables you want the agent to access. Configure which columns are searchable and which should be returned in responses.

Pro tip: Create a Dataverse view specifically for your agent that includes only the columns it needs. This improves response quality and reduces token usage.

Custom API Knowledge Sources

Best for: Real-time data from external systems - CRMs, ERPs, weather, inventory

If your data lives outside Microsoft 365, use a custom connector or Power Automate flow as a knowledge source.

Setup: Create a Power Automate flow with an HTTP trigger, build your API call logic, and register it as a plugin action in Copilot Studio.

Copilot Studio -> Plugin Action -> Power Automate -> External API -> Response

Which Should You Use?

Scenario	Knowledge Source
Employee handbook Q&A	SharePoint
Order status lookup	Dataverse
Real-time inventory check	Custom API
IT troubleshooting guide	SharePoint
Customer account details	Dataverse + Custom API

In practice, most production agents combine two or more knowledge source types.