Data connections
A data connection links the workspace to an external data source so agents can use it. It's the foundation a semantic data model builds on; data frames can come from a connection's query results too (or from files you upload).
Supported sources
Databases and warehouses — PostgreSQL, Redshift, Snowflake, MySQL, BigQuery, and Databricks.
How it works
- A connection is defined once for the workspace, with its host and credentials.
- You then attach the connections an agent is allowed to use, so each agent only reaches the data it should.
- An agent never connects to a source that hasn't been attached to it.
Validate the connection
When you create or edit a connection, use Validate connection to test it before saving — the platform connects with the details you entered and reports back. If validation fails, nothing is saved, so you fix the host, port, or credentials and try again rather than discovering the problem when an agent runs.
Good to know
- Database connections egress to the source on its own port (often not 443). On self-hosted, allow that egress — see Networking rules.
- What an agent can do with a connection — read-only, or specific write actions — is controlled on the semantic data model through operations.