Databricks AI/BI Genie is a conversational analytics interface that lets business users ask questions in plain English and receive answers backed by SQL queries on Unity Catalog data. Unlike simple chatbots, Genie uses a compound AI system — combining language models, semantic metadata, prompt matching, and query verification — to generate and validate SQL against your actual data. It became generally available (GA) on June 12, 2025, and runs entirely within the Databricks platform governance model, meaning Unity Catalog row filters and column masks apply automatically to every query Genie generates.
What This Cheat Sheet Covers
This topic spans 14 focused tables and 105 indexed concepts. Below is a complete table-by-table outline of this topic, spanning foundational concepts through advanced details.
Table 1: Core Concepts and Architecture
Genie is not a single model but a system of cooperating components. Understanding its architecture helps you tune it correctly and set realistic expectations about what it can and cannot answer accurately without additional configuration.
| Concept | Example | Description |
|---|---|---|
A "Sales Analytics" space connected to sales.crm.opportunity and sales.crm.accounts | A curated conversational interface built by domain experts (analysts/data engineers) that exposes specific Unity Catalog tables and contextual instructions to business users. | |
NL → SQL generation → Inspect verification → result | • Genie is not a single LLM • it chains multiple AI steps — query generation, verification, result summarization — into a coordinated pipeline | |
Table descriptions + SQL expressions + example queries | • The living semantic model attached to a Genie space • curators add metadata, instructions, and SQL snippets that teach Genie domain meaning | |
Tables must be registered in Unity Catalog to be added to a space | • Genie only operates on Unity Catalog tables and views • all access is governed by existing UC row filters, column masks, and privileges |