Skip to main content

Warehouse connections

Zingle AI runs SQL against your data warehouse to preview data, validate models, estimate cost, and answer questions. This is configured with three related objects.

Connections​

A connection describes how to reach a warehouse - the provider (Snowflake), account, and connection-level configuration. It's created once by an administrator and shared across the organization.

Targets​

A target is a specific compute context within a connection - the database, schema, and warehouse/compute to run against. A single connection can have multiple targets (for example, an analytics warehouse and a smaller ad-hoc warehouse). Targets are what environments bind to.

Credentials​

Credentials are the actual secrets used to authenticate. They are held per user and encrypted at rest, so SQL runs as you, and access follows your warehouse grants. When an agent executes a query, it resolves your credential for the relevant connection.

Testing & browsing​

From the Infra area you can:

  • Test a connection to confirm credentials and reachability.
  • Browse the live warehouse - list databases, schemas, and tables; inspect table columns; and preview rows.

Safety​

All warehouse SQL goes through a hardened execution path:

  • User-supplied identifiers (database/schema/table names) are validated against a strict allowlist and always double-quoted.
  • Values in predicates use bound parameters - never string interpolation.
  • Result sets are capped and queries have server- and client-side timeouts.