Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.summand.com/llms.txt

Use this file to discover all available pages before exploring further.

Connectors link Summand to a live data source so analysis stays current without manual re-uploads. Pick a table, hit analyze, and Summand handles ingestion, semantic layer computation, and surprise finding end-to-end.

​
Supported sources

  • Databricks (Delta Sharing) — connect via a Delta Sharing credential profile to read tables out of a Databricks Unity Catalog share.
  • Azure SQL — host/port/database connection to an Azure SQL server with SQL or Entra credentials.
  • Snowflake — account-based connection with warehouse and credential selection.
  • CSV upload — one-shot file upload for ad-hoc analysis without a live source.

​
Databricks (Delta Sharing)

Summand connects to Databricks through the open Delta Sharing protocol — no JDBC driver, no cluster required. Provide the credential profile (the JSON file Databricks generates when you create a recipient) and Summand will:
  1. List shares available to the credential.
  2. List schemas under the chosen share.
  3. List tables under the chosen schema.
  4. Read the selected table on every refresh.
Delta Sharing traffic comes from Summand’s two static egress IPs, so a Databricks IP access list only needs to allowlist those two addresses.

​
Azure SQL

Azure SQL connectors take a host, port, database, and credentials, then expose the same schema browser and table picker as the other database connectors. Summand reads the table on each refresh; nothing is mutated on your server.

​
What a connector gives you

  • Schema discovery. Browse databases, schemas, and tables before committing to one.
  • Table selection. Pick a single table per dataset; the schema is captured at ingestion time.
  • Validation. Summand sanity-checks the table (row count, column types, target column eligibility) before kicking off a run.
  • Scheduled refresh. Hourly, daily, or weekly recomputation. Scheduled refresh skips ingestion when the source hasn’t changed and re-runs only the semantic layer, which is roughly 80% cheaper than a full reload.
  • Sharing. Connectors can be shared with teammates so they can build datasets off the same source without re-entering credentials.

​
Lifecycle

Create connector → Browse schema → Pick a table → Trigger analysis
                                                    ↓
                                         Semantic layer computes
                                                    ↓
                                         Surprise Finder runs
                                                    ↓
                                       Dataset is ready in dashboard
You can edit a connector’s configuration, rotate credentials, or delete it from the connector detail page. Deleting a connector cascades to any datasets built from it.