What Is a Fabric Data Agent?
A Fabric data agent is a conversational AI item in Microsoft Fabric that answers plain-language questions from governed sources, such as a Power BI semantic model, and can be used inside Microsoft 365 Copilot. You connect it to a small set of data sources, give it instructions about your business, and people ask it questions in plain language. The agent decides which source fits the question, writes the query (DAX for a Power BI semantic model, SQL for a lakehouse or warehouse), runs it and explains the result.
Data agents aren't limited to Fabric. Microsoft lets you publish one to Microsoft 365 Copilot, use it from Copilot Studio, or reach it through an MCP endpoint from other AI tools. That reach is why data agents matter to analytics teams: one agent, built on one set of definitions, can sit behind a chat in Microsoft 365 Copilot and behind a question asked from another tool.
How a Fabric Data Agent Answers a Question
- It reads the question and its instructions. Agent instructions cover routing and tone, such as which source to prefer for financial questions. For a Power BI semantic model, Microsoft says business terms belong in the semantic model's own AI instructions, because the agent ignores agent-level instructions when it writes DAX.
- It picks a source. Microsoft limits each agent to five data sources, so most teams scope an agent to one subject area, such as finance or sales.
- It writes a query. Against a Power BI semantic model it writes DAX that uses the measures and relationships defined there. Against a lakehouse it writes SQL.
- It runs the query and explains the result. The answer comes back in plain language, and the query behind it can be reviewed.
Why the Semantic Model Decides the Quality of the Answer
A data agent doesn't know your business. It knows what its sources tell it. Pointed at raw ERP tables, it sees column names and codes, not which fields roll up to Total Sales or which order types count as revenue. Pointed at a governed Power BI semantic model, it starts from measures that carry those rules, so its job shrinks from inventing a calculation to choosing the right one.
That's why Microsoft's Prep data for AI settings matter so much here. AI instructions, the AI data schema and verified answers are stored on the semantic model, so every report and every data agent built on it inherits them. The full list is in our definition of an AI-ready semantic model.
Fabric Data Agents and Security
When a data agent queries a Power BI semantic model, the row-level security defined on that semantic model applies to the person asking, so people only see what they're allowed to see. QuickLaunch has tested this in Fabric data agents, Microsoft 365 Copilot Chat and the Power BI Service. One Power BI rule to know before rollout: people with edit rights on a semantic model aren't subject to its row-level security, so test with a real restricted user.
What It Takes to Run One
Fabric data agents run on a paid Fabric capacity, and using them inside Microsoft 365 Copilot has its own licensing requirements. Neither is free, and Microsoft's requirements change as features leave preview, so check Microsoft's current documentation before you plan a rollout.
Frequently Asked Questions
What is a Fabric data agent used for?
It lets people ask questions of governed data in plain language and get an answer with the query behind it. Teams use data agents for subject areas such as finance, sales or operations, and publish them where people already ask questions, including Microsoft 365 Copilot.
Can a Fabric data agent use a Power BI semantic model?
Yes. A Power BI semantic model is one of the sources a data agent can use, along with lakehouses, warehouses and other Fabric sources. With a semantic model as the source, the agent answers from the measures and relationships defined there.
What's the difference between a Fabric data agent and Microsoft 365 Copilot?
Microsoft 365 Copilot is the AI assistant people use across Microsoft 365. A Fabric data agent is a specialist for a defined set of data. Publish the agent to Microsoft 365 Copilot and people can ask it data questions from the same chat they use for email and documents.
Do Fabric data agents work with ERP data?
They work with whatever sources you connect. For ERP data from systems like JD Edwards, Vista or NetSuite, the agent answers best when its source is a semantic model that carries the ERP's business logic, rather than the raw tables.
Fabric Data Agents and QuickLaunch's Approach
QuickLaunch Analytics builds on Microsoft's native MCPs and Fabric data agents. Each Application Pack delivers a governed Power BI semantic model for the system you run, from JD Edwards and Vista to NetSuite, OneStream, Salesforce and Spectrum. QuickLaunch semantic models ship with the descriptions and instructions AI tools read, so a Fabric data agent built on one answers from the same trusted definitions as your reports. See how it fits together on our AI page, or read what AI needs from your data.
More plain-English definitions of data, analytics and AI terms are in the QuickLaunch Learning Center.