databricks-genie

First inspect your tables with get_table_details to learn column names and types, then call create_or_update_genie with a display name, table identifiers, a description of table relationships, and sample questions referencing actual columns.

Use the ask_genie tool with a space_id and your natural language question. It returns the generated SQL, columns, data rows, and a conversation_id you can pass to ask_genie_followup for context-aware follow-up questions.

Use ask_genie when the space has curated business logic, certified queries, or the user explicitly requests Genie. Use execute_sql for simple ad-hoc queries, when you already have exact SQL, or when no Genie Space exists for the data.

Prefer silver and gold layer tables since they are cleaned and optimized for analytics; avoid bronze raw data. Include related tables that join together, use descriptive column names, and add table comments so Genie understands the data.

Poor query generation usually comes from vague column names, missing table comments, or a lack of sample questions. Add instructions and sample questions that reference actual column names, and curate the space further in the Databricks UI.

If warehouse_id is not specified, the tool auto-detects warehouses, preferring running ones and smaller sizes. If none exist, it returns an error, so you must create a SQL warehouse or provide a warehouse_id explicitly.