Microsoft Fabric continues to deliver powerful innovations—and Fabric Data Agents might just be one of the most exciting features to come out of FabCon. If you’re looking to build conversational experiences directly on top of your Power BI and Fabric data, this is your moment.
In this blog post, we’ll explore what Fabric Data Agents are, how to build one step-by-step, and how to customize and deploy it for dynamic, user-friendly Q&A across your data ecosystem.
What Are Fabric Data Agents?
Fabric Data Agents allow you to create natural language chat interfaces that query your data—semantic models, lakehouses, and more. While tools like Copilot offer similar capabilities, data agents give you more control and customization over how users interact with your datasets.
You can define:
- Data sources and tables the agent can access
- How column names and metadata are interpreted
- Example questions and the logic behind them (via SQL or DAX)
- Access permissions for security
Think of it as creating a specialized, AI-powered analyst for your team.
Setting Up a Fabric Data Agent: Step-by-Step
1. Navigate to Microsoft Purview and Create Your Agent
Start by heading into the Microsoft Purview hub and selecting “Create Data Agent.” Give your agent a meaningful name—this will represent your domain of knowledge.
2. Connect to a Data Source
Choose your primary source. In the example, Marthe connected to a semantic model called Metrics Analytics, which includes fact and dimension tables for “Guy in a Cube’s” YouTube metrics.
Add only the relevant tables (e.g., video stats, sentiment analysis data). You can customize further by selecting columns like:
Author: the user who commentedLikeCountandReplyCount: specific to YouTube comment interactions
3. Add Metadata and Hints
Help your data agent understand your data better:
- Label important relationships (e.g., Author = Commenter)
- Define how columns should be interpreted
- Suggest which tables should answer specific queries
The more semantic clarity you provide, the smarter the agent becomes.
4. Define Example Queries
You can add sample questions and the SQL/DAX logic needed to answer them. For example:
- Question: “What host is most popular based on average sentiment score?”
- Query: A DAX or SQL expression fetching and sorting sentiment values by host.
This improves performance and ensures accurate responses.
Testing the Data Agent
Once set up, test your agent by asking real questions like:
- “Total views for January 10, 2025?”
- “Which title has the most views that includes ‘Fabric’?”
The agent responds using either DAX (for semantic models) or SQL (for lakehouses). You can even inspect the underlying queries it generates to validate accuracy.
Results might round or approximate values—validate through Power BI visuals for consistency.
Sharing and Permissions
You can share your data agent with other users, like team members or executives. Just remember:
Users must also have access to the underlying data source to query through the agent.
Marthe demonstrated how users like “Justin Case” can seamlessly interact with the agent—asking basic questions like “Who are the hosts?” and receiving quick, accurate responses.
Go Beyond the UI: Programmatic Access
Want even more control? You can use the Fabric Data Agent SDK to interact programmatically using Python. Define your agent, instantiate the client, and run queries like:
"Top 5 videos with Fabric in the title by comments"
This opens doors for enterprise integration, automation, and embedding AI features into your own tools.
Scale with Azure AI
Fabric Data Agents aren’t limited to Fabric. You can extend usage into Azure AI solutions, integrating with Azure AI Agent or Azure Foundry for broader enterprise use.
This makes Fabric Data Agents not just a reporting feature, but a strategic AI tool across your entire data infrastructure.
Final Thoughts
Fabric Data Agents are a game-changer for interactive analytics. They:
- Simplify natural language querying
- Support custom logic
- Integrate across Microsoft Fabric and Azure
If you’re thinking of trying it, start small—connect your semantic model, test a few queries, and iterate.
Want to level up your Power BI and Fabric skills?
Explore expert-led training here:
