Databricks Genie

Databricks Genie is the conversational front end to data that already lives in Databricks. A business user asks a question in plain English; Genie generates SQL against a curated set of Unity Catalog tables, runs it on a warehouse, and returns a result the asker can see the query behind. The governance is not bolted on: Unity Catalog permissions, row filters, and column masks apply to the agent exactly as they apply to a human running the same query.

The naming moved in 2026, so read old material carefully. Genie started as a conversational analytics assistant in Databricks AI/BI. On June 16, 2026 Databricks introduced Genie One (the business-user experience), Genie Agents (what Genie spaces became), and Genie Ontology (an automatic context layer that extracts knowledge from tables, queries, dashboards, and pipelines into a graph). Genie Code is the developer-facing side of the same family.

Highlights

• Grounded in Unity Catalog. A Genie Agent is scoped to Unity Catalog data, up to 30 tables or views, and inherits the catalog's access policies, row filters, and column masks. Nothing the asker could not query themselves comes back.
• Curated context, not raw text-to-SQL. Subject-matter experts supply general instructions, example SQL queries, and common questions, so the agent learns your joins and your definitions rather than guessing at them.
• Trusted answers via SQL functions. Predefined functions with parameterized SQL let you pin the exact query behind a high-stakes question, so it returns a verified answer every time instead of a freshly generated one.
• Metric views as the shared semantics. Unity Catalog metric views separate measure definitions from the dimensions used to group and filter them, and can be queried from SQL editors, notebooks, dashboards, Genie Agents, and alerts. Genie can also export an agent's context into a metric view.
• An API for your own surfaces. Chat mode APIs let an application, chatbot, or agent framework start a stateful conversation, ask follow-ups, and retrieve the generated SQL, the query results, and the visualizations.
• Feedback loops built in. Activity monitoring tracks the questions asked, usage, and user ratings, which is how you find the definitions your instructions are still missing.

In an analyst's workflow

Genie rewards preparation more than prompting. The setup work is where the accuracy comes from:

-- Register the answer you never want re-derived:
CREATE FUNCTION analytics.finance.net_revenue_by_month(
  start_date DATE, end_date DATE
)
RETURNS TABLE
RETURN
  SELECT date_trunc('MONTH', invoice_date) AS month,
         SUM(amount) - SUM(refund_amount) AS net_revenue
  FROM analytics.finance.fct_invoices
  WHERE invoice_date BETWEEN start_date AND end_date
  GROUP BY 1;

Add that function to the agent, write an instruction saying net revenue always means this, and "what was net revenue in July?" stops being a coin flip. Then watch the activity feed: every low-rated answer is a missing instruction or a missing example query. For the same grounding problem outside a single vendor's platform, see text-to-SQL with Claude and the best text-to-SQL tools in 2026.

Good to know

Genie is part of the Databricks platform rather than a separate purchase, and it is billed by consumption. As of September 2026 Databricks documents that Genie One and Genie Agents usage by users is free until January 31, 2027 and pay-as-you-go after that, while service principal traffic is billed the whole time; Genie Code moved to pay-as-you-go on July 8, 2026, with 150 free DBUs per user per month and overage on the Serverless Real-Time Inference SKU. The compute your queries run on is billed separately either way. Your data must be managed in Unity Catalog, and the feature set assumes Pro or Serverless SQL compute.

That makes the real comparison a platform one. If your governed semantic layer lives in Databricks, Genie is the cheapest good answer. If it lives elsewhere, or if you want the same conversational surface over several stacks, ThoughtSpot Spotter is the closest dedicated competitor and Hex is the closest notebook-plus-agent alternative. The best AI tools for data analysts in 2026 roundup covers how they differ in practice, and Claude for data analysis covers the general-assistant route.