← All guides
Connectors

Connect Databricks to Saemi

Step-by-step guide to connect a Databricks workspace to Saemi using a personal access token. Covers workspace URL, warehouse selection, and catalog scope.

Last updated: 2026-06-14

Saemi runs your questions through Databricks SQL warehouses using a Personal Access Token (PAT). The connection takes about two minutes.

Before you start

You need:

  • A Databricks workspace URL (e.g. https://dbc-12345678-abcd.cloud.databricks.com)
  • A running SQL warehouse (Saemi runs queries through it — Serverless is fine, classic warehouses work too)
  • Permission to create a Personal Access Token

Step 1 — Create a Personal Access Token

  1. In Databricks, click your profile avatar (top right) → User Settings
  2. Open the Developer tab → Access tokens → Manage
  3. Click Generate new token
  4. Set a comment ("Saemi production"), lifetime (90 days is reasonable — Saemi will warn you on expiry)
  5. Copy the dapi… token — Databricks shows it only once

Step 2 — Find your warehouse HTTP path

  1. SQL Warehouses (left nav) → click your warehouse
  2. Connection details tab
  3. Copy the HTTP Path (e.g. /sql/1.0/warehouses/abc123def456)

Step 3 — Connect in Saemi

  1. Open /accounts/connector/ in Saemi
  2. Click Databricks → Connect
  3. Paste:
    - Workspace URL — without trailing slash
    - HTTP Path — from step 2
    - Personal Access Token — from step 1
  4. (Optional) Catalog scope — restrict Saemi to specific catalogs (e.g. main, gold). Leaves other catalogs invisible. Recommended for teams with PII catalogs they don't want the LLM to touch.
  5. Click Test connection — Saemi runs SELECT 1 and reports back
  6. Click Save

After save, Saemi scans the schema (about 30-90 seconds) and builds an ERD of your top tables. You'll see it in your new chat home.

Permissions checklist

For the PAT to work, your Databricks user needs:

  • CAN_USE on the SQL warehouse
  • SELECT on the catalogs/schemas you want Saemi to read
  • (Optional) USE_CATALOG on each catalog if Unity Catalog is enabled

Saemi only runs read queries — it never issues INSERT, UPDATE, DELETE, or DDL against your warehouse.

Troubleshooting

"Connection refused" — check the workspace URL (HTTPS, no trailing slash) and that the SQL warehouse is running (not Auto-Stopped).

"Permission denied on schema X" — your PAT user lacks SELECT on that schema. Either grant it or narrow the catalog scope.

"Token expired" — Databricks tokens have a max lifetime. Regenerate and re-save in Saemi.

What Saemi sees

Saemi reads:

  • Table and column names (for autocomplete + ERD)
  • Row counts (to suggest sampling strategies)
  • Query results (for the answers it returns)

Saemi does NOT store your data — query results live only in the chat row, encrypted at rest, and are deleted when you delete the chat.