Microsoft Fabric Data Agents - Conversational AI for Data Analytics#
Microsoft Fabric Data Agents represent a revolutionary new feature that enables the creation of conversational Q&A systems using generative AI. These agents make data insights accessible and actionable for everyone in the organization - even for users without deep technical AI knowledge or understanding of data structures.
What are Fabric Data Agents?#
Data Agents in Microsoft Fabric enable teams to interact with their data in natural language. Users can ask questions in plain English/German and receive precise, context-rich answers from data stored in Fabric OneLake.
Core Capabilities:#
- 🗣️ Natural Language Processing: Ask questions in natural language
- 🔍 Intelligent Data Source Selection: Automatic identification of the most relevant data source
- 🛡️ Secure Access Control: Uses user credentials for permissions
- 📊 Multi-Query Support: SQL, DAX, and KQL depending on the data source
- 🎯 Organization-Specific Customization: Custom instructions and example queries
How do Data Agents Work?#
The Fabric Data Agent uses Azure OpenAI Assistant APIs and operates in a structured process:
1. Question Parsing & Validation#
- Processing user questions through Azure OpenAI
- Compliance with security protocols and Responsible AI policies
- Read-only access to all data sources
2. Data Source Identification#
- Using user credentials for schema access
- Evaluation of all available data sources:
- Lakehouse & Warehouse (relational databases)
- Power BI Datasets (Semantic Models)
- KQL Databases
3. Tool Invocation & Query Generation#
- Natural Language to SQL (NL2SQL) for Lakehouse/Warehouse
- Natural Language to DAX (NL2DAX) for Power BI Datasets
- Natural Language to KQL (NL2KQL) for KQL Databases
4. Query Validation & Execution#
- Validation of generated queries
- Secure execution against the selected data source
- Formatting into human-readable answers
Data Agent Configuration#
Supported Data Sources#
- Up to 5 data sources in any combination
- Lakehouses, Warehouses, KQL Databases, Power BI Semantic Models
- Extended support for large data sources (>1,000 tables, >100 columns)
Customization Options#
Data Agent Instructions:
- Guidance for selecting the best data source
- Organization-specific rules and definitions
- Example: “Financial metrics → Power BI Semantic Model”
Example Queries:
- Example question/query pairs
- Improved accuracy through context
- Currently not available for Power BI Semantic Models
Integration & Usage#
Consumption Options:#
- Microsoft Copilot Studio (Connected Agents)
- Azure AI Foundry (Python SDK, REST API)
- External Applications (Python Client SDK)
- Microsoft Teams and other channels
Prerequisites:#
- Fabric Capacity F2+ or Power BI Premium P1+
- Enabled Tenant Settings for Data Agents and Copilot
- Cross-geo Processing/Storing for AI enabled
- At least one data source (Warehouse, Lakehouse, Power BI, KQL DB)
Current Developments (2025)#
New Features:#
- July 2025: Extended support for large data sources
- June 2025: Data Source Instructions for more precise answers
- September 2025: Python Client SDK for external applications
Limitations (Preview):#
- Only “Read” queries (no Create/Update/Delete)
- No unstructured data (.pdf, .docx, .txt)
- Currently only English language optimally supported
- No switching of underlying LLM possible
Further Resources#
Concepts & Fundamentals:#
Implementation:#
Integration:#
- Use with Azure AI Foundry (Python)
- Use with Microsoft Copilot Studio
- Python Client SDK for External Apps
Scenarios & Use Cases:#
Conclusion: Microsoft Fabric Data Agents democratize access to data insights through conversational AI and create the foundation for a data-driven corporate culture. With integration into various Microsoft platforms and continuous development, they offer enormous potential for the future of data analytics.