Configure your Power Apps Copilot to talk with your Dataverse data | DEM525

by Alex Rostan

Day 2 · Demo Session · 14 min · 3 min read

Applied AI

TL;DR

Alex Rostan demonstrated how to wire a Copilot Studio agent into a Power Apps canvas app in under two minutes: create an agent, connect it to a Dataverse Accounts table, publish, and select it in Power Apps settings. The same agent then answered natural-language questions in Spanish and English against live data, returning City Power of Light as the most profitable company in Redmond with $100,000 in revenue. A second pass added SharePoint documents and a public website as knowledge sources, showing one agent can span three data tiers without duplication.

Power Platform Components and the Copilot Layer

▶ Watch (01:41)

Microsoft Power Platform bundles four tools: Power BI for analytics, Power Apps for custom applications, Power Automate for business flows, and Copilot Studio for building agents. The platform’s stated goal is letting individuals and teams build applications and automate workflows with minimal code. Copilot Studio is the piece this session focuses on. It sits alongside Power Apps rather than inside it, which matters because one agent configured there can be attached to any Power App you publish.

Creating and Connecting a Copilot Agent in Copilot Studio

▶ Watch (04:23)

From Copilot Studio’s homepage, create a new agent and add a knowledge source. Select “database,” search for the target Dataverse table (in the demo, the Accounts table), and click Next. That connects the agent to your data. Then hit Publish. Back in Power Apps, open Settings, go to the Copilot tab, pick the agent you just published, and save. The app now carries an embedded copilot that can answer questions about whatever table you connected. The whole sequence takes under two minutes.

Querying Dataverse Data Through the Copilot in Natural Language

▶ Watch (07:34)

With the agent live inside the canvas app, Rostan ran the live Dataverse query demo (07:34): “which is the most profitable company in Redmond?” The copilot returned City Power of Light, matching the table record showing $100,000 in revenue. The same query works in Spanish, French, or Portuguese with no extra configuration. The agent interprets the question regardless of language and returns the same structured result.

Extending the Copilot with SharePoint and External Web Sources

▶ Watch (10:31)

The same agent that queries Dataverse can also pull from SharePoint document libraries and public websites, all without building a separate agent. Rostan added a SharePoint site holding product and supplier documents as a second knowledge source. The SharePoint and web source demo (11:40) asked where Rainforest Exotic Goods Supplies is located; the copilot returned “San Pablo, Brazil” and linked back to the source Word document. He also added www.microsoft.com as a third source, showing that any REST API, internal system, or public portal can feed the same agent.

Notable Quotes

Demo data about companies, revenues, locations, and numbers. Alex Rostan · ▶ 3:14

You can ask on different language. Alex Rostan · ▶ 8:28

We can integrate everything on the same compiler. Alex Rostan · ▶ 13:51

Key Takeaways

One Copilot Studio agent connects to Dataverse, SharePoint, and public websites without building separate agents for each.
Configuring the agent requires three steps: connect to a table, publish, then select the agent in Power Apps settings.
Multilingual queries work out of the box; no additional configuration is needed for Spanish, French, or Portuguese.

Filed under

Applied AI