Connect Databricks Genie to Glean Assistant
Glean Assistant can query Databricks data using natural language through Databricks Genie spaces. When a user asks a data question in Assistant, Glean identifies the best-matching Genie space from indexed metadata and routes the question to Databricks Genie for execution. Databricks SQL is available as an explicit fallback when no Genie space matches.
How it works
Unlike direct SQL integrations, the Databricks experience in Assistant is Genie-first. Glean's Databricks connector crawls and indexes your Genie spaces, including titles, descriptions, sample questions, and table metadata. When a user asks a question, Assistant uses this indexed metadata to select the most relevant Genie space and sends the natural-language question directly to Databricks Genie.
Databricks SQL is not the primary query path. Assistant uses SQL only when a user explicitly requests it or when no suitable Genie space can be found.
Prerequisites
Before you begin, ensure you have the following:
- Administrative privileges in both Glean and Databricks.
- The Databricks connector configured and actively indexing Genie spaces. This is required so that Assistant can discover and route to the correct Genie space.
- The Databricks Genie spaces you want to expose are properly configured in Databricks with descriptive titles, descriptions, and sample questions.
Set up Databricks Genie in Assistant
The setup has two main parts: configuring the Databricks action pack (if not already done) and enabling the actions in Assistant.
Step 1: Set up the Databricks action pack
If you haven't already configured the Databricks action pack, follow the instructions in Databricks actions setup. The action pack includes two actions:
- Search Databricks with Genie — routes natural-language questions to a Genie space.
- Search Databricks with SQL — runs direct SQL queries against a Databricks warehouse.
The action pack setup involves creating an OAuth app in Databricks and registering the credentials in Glean.
Step 2: Verify Databricks connector indexing
Ensure that your Databricks connector is indexing Genie spaces. Assistant relies on the indexed Genie space metadata to identify which Genie space can best answer a user's question.
- Navigate to Admin console → Connectors.
- Locate the Databricks connector and verify it shows a healthy status.
- Confirm that Genie spaces appear in Glean search results.
If your Genie spaces don't appear in search, verify that the connector has completed at least one full crawl and that the Genie spaces have proper permissions configured in Databricks.
Step 3: Enable Databricks in Glean Assistant
After the action pack and connector are configured, enable the Databricks actions for Assistant:
- Navigate to Admin console → Platform → Actions.
- Search for the Databricks action pack.
- Under Configuration, go to the Enable actions section.
- Click Edit settings.
- Click Chat and define visibility settings to choose who can use Databricks in Glean Assistant and agents.
Step 4: Test the integration
Verify the setup by asking Glean Assistant questions about your data.
Example prompts:
- Natural language: "What was our revenue by region last quarter?"
- Direct SQL: "Run this SQL against Databricks:
SELECT region, SUM(revenue) FROM sales GROUP BY region;"
For a natural-language question, Assistant should indicate that it used Databricks Genie and show which Genie space it selected.
How Genie space routing works
When a user asks a data question, Assistant performs the following steps:
- Determines relevance — decides whether Databricks is the right data source for the question.
- Selects a Genie space — searches the indexed Genie space metadata (titles, descriptions, sample questions, and table information) to find the best match.
- Sends the question — forwards the natural-language question to the selected Genie space via the Databricks Genie API.
- Returns results — displays the Genie response, including any generated tables, directly in Assistant.
If no Genie space matches, Assistant falls back to Databricks SQL only when the user explicitly requests a SQL query.
Permissions and access control
Data access in Assistant follows your existing Databricks permissions:
- Users authenticate with their own Databricks credentials via OAuth.
- Query results respect the permissions configured on Genie spaces and underlying tables in Databricks.
- The Databricks connector enforces permission-aware access to indexed Genie space metadata, so users only see Genie spaces they have access to.