Databricks Genie — text-to-SQL agent

What it is

Databricks Genie is a governed natural-language query agent built into Databricks SQL. A business user types a question in plain English and receives a SQL-generated answer against Delta Lake tables, without writing SQL themselves. It is deliberately not a general-purpose chat interface: every Genie deployment is scoped to a Genie Space — an explicit configuration of the Delta tables in play, a semantic layer of business-term definitions, curated sample questions, and trusted SQL snippets that the data team authors and maintains. The large language model behind Genie draws on this curated context rather than open-ended reasoning about arbitrary tables, which is what keeps its answers close to the ground truth the data team has actually validated.

Why it matters: for organisations that have already invested in bringing SAP data into Databricks — typically through SAP Business Data Cloud's Delta Sharing integration — Genie is the fastest way to convert that replicated data into genuine self-service for non-technical stakeholders. A finance or supply-chain analyst who cannot write a join across three fact tables can still ask what margin looked like by product line in Germany last quarter and get a defensible answer, with the underlying SQL exposed via a View SQL button for anyone who wants to audit exactly what ran. This closes a long-standing gap in enterprise BI: dashboards answer the ten questions someone anticipated, while Genie answers the eleventh question nobody built a report for.

Why it matters

• Genie only works within its configured 'Genie Space' — a fixed set of Delta tables, business-term definitions, and trusted SQL snippets, not open internet knowledge.
• It's the natural fit for repetitive, well-documented reporting: finance spend by cost centre, sales pipeline by region, supply-chain inventory by plant.
• When SAP data reaches Databricks via BDC Delta Sharing, a Genie Space over those tables lets finance or supply-chain teams self-serve without SAP licences or SQL skills.

Key points

• Genie is a governed natural-language query agent inside Databricks SQL: a business user asks in plain language and receives a SQL-generated answer over Delta tables.
• Every deployment is scoped to a Genie Space — explicit tables, business-term definitions, sample questions and trusted SQL snippets authored by the data team.
• The governance guarantee: Unity Catalog enforces the user's own permissions at query time, so there is no privilege escalation through the natural-language interface.
• Answer quality follows the curation of the Space, not the model — an unmaintained Space drifts as fast as the schema does.
• AI/BI Genie is generally available at no additional licence cost to Databricks SQL customers, with every AI/BI Dashboard now including an integrated Genie space.
• Genie Conversation APIs let a Genie Space be embedded into Slack, Microsoft Teams or a custom application — mirroring the collaboration-tool integration SAP offers for Joule Work.
• Genie is exposed as a managed MCP server, meaning it can in principle be called as a tool by any MCP-compatible orchestrator, including a Joule agent built via the A2A 'Bring Your Own Agent' pattern.
• For SAP-originated data, Genie's semantic layer does not inherit SAP process context (fiscal calendar, SCD-current-record definitions) — that has to be declared explicitly, not assumed.

Terms used on this page

Genie Space

The scoped configuration unit of a Genie deployment: explicit tables, business-term definitions, sample questions and trusted SQL snippets authored and maintained by the data team.

View SQL

The button Genie exposes on every answer, showing the exact generated SQL — the mechanism that lets the answer be audited rather than taken on faith.

Genie Conversation API

Databricks' 2026 API for embedding a Genie Space's conversational interface directly into Slack, Microsoft Teams, or a custom internal application.

Managed MCP server (Genie)

Databricks' MCP-compatible exposure of a Genie Space, allowing any MCP-aware orchestrator — including a Joule agent — to call it as a tool.

SCD Type 2

A slowly-changing-dimension pattern that keeps both historical and current versions of a record; a Genie Space over such a table must explicitly define what counts as the 'current' row or it will double-count.

Semantic layer (Genie vs Datasphere entity model)

Genie's semantic layer is a lightweight, separately authored set of business-term definitions and SQL examples over Delta tables; Datasphere's entity model is both the data model and the semantic layer for Joule's NL query, since one artefact serves both purposes.

Sources

1. Databricks — Genie text-to-SQL docs
2. Databricks Blog — AI/BI Genie is now Generally Available
3. Databricks Docs — Genie overview (AWS)
4. Databricks Docs — Genie Spaces with dashboards (GCP)
5. Databricks API Reference — Genie Space object
6. Databricks Blog — The next generation of Databricks Genie
7. Databricks Blog — From Data to Dialogue: best-practices guide for building Genie Spaces
8. Databricks Devhub — Genie Agents documentation
9. Databricks SDK for Python — Genie workspace API reference
10. Databricks Docs — AI/BI and Genie One release notes 2026 (Genie One memory GA, 29 Sep; Oct 1 Genie One updates)
11. Databricks Docs — Tune Genie agent quality (knowledge store, example queries, limits)
12. Databricks Docs — Test and monitor a Genie agent (benchmarks up to 500 questions)
13. Databricks Docs — Create and manage a Genie agent (50-table limit, per-user Unity Catalog permissions, certification)
14. Databricks Docs — Use the Genie agents API (chat mode, agent mode, management APIs)
15. Databricks Docs — October 2026 platform release notes (cross-engine ABAC GA, 1 Oct)