Microsoft Fabric Updates Blog

New in Fabric Data Agent: Data source instructions for smarter, more accurate AI responses

We’re excited to introduce Data Source Instructions, a powerful new feature in the Fabric Data Agent that helps you get more precise, relevant answers from your structured data.

What are Data Source instructions?

When you use the Data Agent to ask questions in natural language, the agent must determine which data source to use and how to query it. Previously, all guidance was scoped at the global level, which worked—but often left room for ambiguity, especially when working with complex schemas or multiple tables.

Data Source Instructions let you define targeted guidance per data source—so you can tailor how the agent interacts with each dataset. Think of it as giving the agent a playbook for each source: which tables to prioritize, what filters to apply, how to join tables, or how to interpret column values.

How it Works

Each data source in your agent configuration now has its own Instructions field. These instructions are used during query generation and improve the agent’s ability to:

  • Select the right tables and columns.
  • Understand data-specific logic (e.g., fiscal calendars, regional codes).
  • Apply consistent filters or transformations.
  • Interpret user intent more accurately.

This is especially useful when your organization has:

  • Custom business rules that vary by dataset.
  • Shared naming conventions across tables (e.g., date, region, status).
  • Specific join logic or default filters that should always apply.

What Should You Include?

Effective instructions are:

  • Clear: Describe when and how certain tables or filters should be used.
  • Consistent: Use standard terms across sources and align with your schema labels.
  • Contextual: Include examples of typical queries or values if needed.

Examples:

  • ‘When asked about historical sales, use the Orders table. Filter using OrderDate.’
  • ‘Always join Sales with Products on ProductID before aggregating.’
  • ‘State codes should match abbreviations (e.g., CA, NY), not full names.’

Get Started

To learn how to configure instructions for your data sources, visit our setup guide:
Configure Your Data Agent with Data Source Instructions

For tips on writing effective instructions, check out our best practices:
Best Practices for Agent Instructions


With Data Source Instructions, you can provide the agent with clearer guidance on how to work with each dataset, which helps improve accuracy and consistency in the responses. Give it a try and see how it improves your results.

Related blog posts

New in Fabric Data Agent: Data source instructions for smarter, more accurate AI responses

January 12, 2026 by Tom Peplow

In October 2025, we introduced OneLake diagnostics—a powerful capability that helps teams “answer who accessed what, when, and how” across your Fabric Lakehouse environment. OneLake diagnostics streams JSON-based activity logs into a Lakehouse you choose, enabling rich analysis, governance, and compliance workflows. A powerful capability that helps teams “answer who accessed what, when, and how” … Continue reading “Gain even more trust and compliance with OneLake diagnostics immutability (Generally Available)”

January 8, 2026 by Adi Eldar

What if generating embeddings in Eventhouse didn’t require an external endpoint, callout policies, throttling management, or per‑request costs? That’s exactly what slm_embeddings_fl() delivers: a new user-defined function (UDF) that generates text embeddings using local Small Language Models (SLMs) from within the Kusto Python sandbox, returning vectors that you can immediately use for semantic search, similarity … Continue reading “Create Embeddings in Fabric Eventhouse with built-in Small Language Models (SLMs)”