Unlocking Data Conversations with Microsoft Fabric Data Agents

Why Fabric Data Agents Matter

Fabric Data Agents are AI-powered conversational assistants embedded within Microsoft Fabric that enable users to ask natural language questions about enterprise data – lakehouses, warehouses, semantic models in Power BI, or KQL databases–and receive insightful answers without writing SQL, DAX or KQL queries. 

They lower technical barriers, empowering stakeholders across your organization–from analysts to executives – to interact with data directly and securely. 

How Fabric Data Agents Work

  1. Multi-Source Support
  • Each agent can connect to up to five data sources, combining lakehouses, warehouses, Power BI semantic models, or KQL databases as needed. 
  1. Natural Language to Queries
  • Uses Azure OpenAI Assistant APIs to parse questions, determine appropriate data sources, and generate NL2SQL, NL2DAX, or NL2KQL automatically based on schema and context. 
  1. Instruction & Example Guidance
  • Supply custom instructions (up to ~15,000 characters) to shape behavior and few-shot example queries to improve accuracy and align responses with business context  
  1. Execution Workflow
  • The agent validates generated queries, executes them (read-only), and formats both intermediate steps and results in conversational output – boosting transparency and trust. 

End to End: Creating Your First Fabric Data Agent

Prerequisites (June/July 2025): 

  • Fabric capacity: F2 or above (not available with Fabric trials). 
  • Tenant switches enabled: Data Agents, Copilot, Crossgeo AI store/processing 
  • At least one supported data source (warehouse, lakehouse, Power BI or KQL) 

Steps: 

  1. In your Fabric workspace, click + New Item → Fabric Data Agent. Name it and begin configuration. 
  2. Add data sources via the OneLake catalog-choosing up to five data assets (e.g. AdventureWorks lakehouse). 
  3. Select tables to expose-only the chosen tables are queryable by the agent. 
  4. Add instructions (e.g. route financial queries to specific sources) and example Q&A pairs to guide the AI. 
  5. Test queries like: “Top 5 products by sales in 2024” and inspect intermediate “steps completed” and generated code. 
  6. Publish your agent-write a description, create a published version tied to a URL or artifact ID, and control incremental updates separately from draft improvements. 

Integrating with Azure AI Foundry & Custom Agents

Fabric Data Agents can also serve as knowledge sources for Azure AI Agent Service (Azure AI Foundry). This enables developers to embed data intelligence into broader AI agents or chatbots in Teams, apps, or orchestration systems: 

  • Add an agent’s workspace ID and artifact ID as a knowledge source in Azure AI Agent configuration. 
  • Use identity passthrough (OBO) to ensure queries respect user permissions. 
  • The Azure AI agent orchestrates when to invoke Fabric Data Agents to answer relevant questions, blending Fabric’s insight with broader logic 

Limitations:

  • Preview feature: Fabric Data Agents are still in public preview, and limitations are expected to be addressed in upcoming updates over time. 
  • Readonly queries only: Agents generate only SQL, DAX, or KQL queries for read operations. They cannot perform data creation, updates, or deletions. 
  • Structured data only: Unstructured resources—such as PDF, DOCX, or TXT files—aren’t supported as data sources. These file types cannot be queried via the Fabric Data Agent. 
  • Englishonly input: The agent currently supports English for prompts, instructions, and example queries; nonEnglish inputs may lead to degraded performance. 
  • Model selection fixed: Users cannot choose or swap out the underlying LLM—Fabric manages model selection automatically, and custom model configuration isn’t supported 

Fabric Data Agents aren’t just a fancy new item available in Microsoft Fabric – they’re a pivotal step toward making data truly conversational and accessible. By combining secure, governed data access with natural language AI, they pave the way for broader adoption of analytics across business teams. And when its insight is woven into Azure AI Foundry agents, Copilot studio, Teams, etc., you unlock intelligent, context-aware assistants that can power teams, processes, and systems – making datadriven decisioning feel as natural as having a conversation. 

Pradeep Saminathan   
Program Director – GNextGen @ Kasadara, Inc