Skip to main content
Databricks connects with a personal access token minted inside one workspace. Every call runs against that workspace’s own host, so connect each workspace you want to reach separately - they appear as separate accounts to pick from.

​
What you can do

SQL - run a statement on a SQL warehouse and get its rows back, and fetch the result of a long-running statement by id. Unity Catalog - list catalogs, schemas and tables, and read one table’s columns and types. Jobs and compute - list SQL warehouses and clusters, list jobs, start a job run, and read a run’s state and per-task results. Triggers - none. Use it through the assistant in chat or as an action node in an automation.

​
Get your credential

1

Open your Databricks settings

In Databricks, open your profile menu → Settings → Developer → Access tokens.
2

Generate a token

Click Generate new token, give it a lifetime, and copy it - Databricks shows it once.
3

Note the workspace URL

Copy the URL you sign in at, e.g. https://dbc-1234abcd-5678.cloud.databricks.com.
To narrow what arg.ai can reach, mint the token on a service principal that holds only the Unity Catalog grants you want it to have, rather than on your own account. A token always carries its owner’s permissions, enforced by Databricks.

​
Connect

1

Open Integrations

In arg.ai, go to Settings → Integrations and click Connect on Databricks.
2

Enter the workspace URL and token

Paste the workspace URL and the token into the connect dialog.
3

Save

arg.ai checks the token against that workspace and connects it.

​
Use it

  • In chat: “Which tables are in main.sales, and how many orders shipped late last week?”
  • In an automation: add a Databricks action node that runs a query on a warehouse, and chain later steps off its rows.

​
How a query works

  • Find the schema first. List catalogs, schemas and tables, then read the table definition, so the SQL is written against real column names rather than guessed ones.
  • Pick a warehouse. A statement runs on a SQL warehouse, so list warehouses and pass the warehouse_id. A query against a stopped warehouse starts it, which Databricks bills.
  • Bind values, don’t paste them. Pass values as named parameters (:name) instead of interpolating them into the SQL text.
  • Long queries return an id. If the statement is still running when the wait elapses, you get back pending: true and a statement_id to fetch the result with.
  • Results are the first chunk. Large results come back flagged rather than paged - narrow the query or lower the row limit instead.

​
What it costs

arg.ai charges nothing extra for the integration. Running SQL and starting job runs consume Databricks compute, billed by Databricks against your account.