AI Query

What is AI Query?

AI Query is a new asset type that allows a producer to provision access to an existing Databricks Genie Agent through the Harbr UI. Users are able to register a Genie Agent they've already built, complete with the tables, instructions, and any pre-existing worked examples.

Creating an AI Query asset

A producer creates an AI Query asset by selecting the AI Query asset type, choosing a Databricks connector and selecting from the Genie Agents that connector can reach, then publishing it like any other asset.

• No credentials are shared with Harbr beyond the connector's own access.

• No data is moved onto the platform.

• There is no crawl or schema step — the asset simply references the existing Genie Agent.

Step 1 - Create a Databricks connector (if one doesn't already exist) pointing to the Databricks instance where your Genie Agent is configured. For user guidance on how to create a new Databricks connector, please see Create a Databricks Connector

Step 2 - Start the asset creation process and select AI Query as the asset type.

Step 3 - Select "Databricks" as the connector type.

Step 4 - Choose the connector that points to your Genie Agent from the dropdown.

Step 5 - Select the specific Genie Agent you want to expose, from the dropdown.

Step 6 - Select “Start Set Up” and finish the asset create journey.

Step 7 - Release the AI Query asset.

Consuming an AI Query asset

Consumers with access to the asset, or are subscribed to a product that contains the AI Query asset, ask questions in natural language via a chat interface, all without leaving the Harbr Platform. You can open the AI Query asset place from:

• The product page

• My Collection

• Home pages (if configured by your platform operator)

Answers are returned as text and tables. The conversation is retained for the duration that the tab is open and the user is signed in. The session can be closed and be cleared to start again — there is no server-side conversation history.

Every question is proxied through Harbr using the connector's identity, so no consumer accesses Databricks directly, and each question is recorded for usage and cost visibility.