What data sources can Genie use for answers and analysis?

Summary

• AI/BI Genie answers over data registered in Unity Catalog — managed tables, external tables, foreign tables, views, materialized views, and metric views.
• In chat mode Genie works with structured data; in Agent mode it can also query unstructured files stored in Unity Catalog volumes.
• Building a Genie Agent on metric views grounds answers in governed, deterministic metric definitions, so natural-language questions resolve to consistent results.
• Genie relies on good context — column descriptions and data dictionaries, curated instructions, sample queries, and verified answers — ideally over query-ready, gold-layer tables.
• Every answer is governed by Unity Catalog: users see only the data they are authorized to see, queries are read-only, and interactions are logged for audit.

What data sources can Genie use for answers and analysis?

Databricks AI/BI Genie answers natural-language questions over your organization's data by generating queries against objects registered in Unity Catalog. It is designed to work with governed, query-ready data, and the quality of its answers depends on both the data sources it can reach and the context you provide about them. See the Genie documentation.

Data sources AI/BI Genie can use

Genie can work with data registered in Unity Catalog, including:

• Managed tables
• External tables
• Foreign tables (data federated from external systems and governed through Unity Catalog)
• Views and materialized views
• Metric views

In chat mode, Genie works with structured data and does not answer questions about unstructured content such as PDFs or Word documents. In Agent mode, Genie can also query unstructured files stored in Unity Catalog volumes.

What Genie needs to answer accurately

• Well-annotated data. Genie uses the metadata attached to Unity Catalog objects — column names and descriptions, data dictionaries, and other documentation — to generate accurate responses.
• Metric views for governed definitions. Building a Genie Agent on metric views grounds answers in governed, deterministic definitions that compile into logical queries at runtime, so results stay consistent instead of relying on inferred logic. See Redefining the semantics data layer for the future of BI and AI.
• Instructions and business context. Author-curated instructions, sample queries, business rules, and verified answers explain jargon and domain logic and improve response quality.
• Query-ready, pre-joined data. Genie is designed to work with governed, gold-layer tables; pre-joining related tables into views or metric views reduces complexity and improves results.

Governance and access

Every Genie answer is grounded in data governed by Unity Catalog. Users see only the data they are authorized to see, every query is read-only, and every interaction is logged for audit. See How Databricks Genie democratizes data access.

Getting started

• Register your analytics data in Unity Catalog and identify the query-ready, gold-layer tables Genie should use.
• Add column descriptions, data dictionaries, and curated instructions, sample queries, and verified answers so Genie has business context.
• Where you have standardized KPIs, define metric views and build the Genie Agent on them for consistent, governed answers.
• Create the Genie Agent and let business users ask questions in the AI/BI interface.

FAQs

What data types can Genie query?

Genie works with data registered in Unity Catalog — managed, external, and foreign tables, plus views, materialized views, and metric views. Chat mode uses structured data; Agent mode can also query unstructured files in Unity Catalog volumes.

How do I make Genie's answers more accurate?

Provide good context: column descriptions and data dictionaries, curated instructions, sample queries, and verified answers, ideally over query-ready gold-layer tables. Building the space on metric views grounds answers in governed, deterministic definitions.

Is Genie access governed?

Yes. Answers are governed by Unity Catalog, so users see only the data they are authorized to see, queries are read-only, and every interaction is logged for audit.