Use Postgres

Create a Postgres service from Databases → New database. Attach it to a project to inject DATABASE_URL, or leave it standalone and store the returned URI in the authorized consumers yourself.

Connect through SproutOS#

Applications connect with the standard Postgres URI in DATABASE_URL. The public endpoint is a SproutOS tenant proxy: it resolves the tenant database, checks suspension state, and drops its own elevated role before forwarding the session. Customers do not receive the underlying provider credential.

The connection URI is shown once. If it is lost or exposed, rotate it from the service menu and replace the old value in every environment that uses it.

Use branches for development and agents#

A database branch is a temporary, isolated copy derived from the service's primary branch. Use one for schema experiments, migration tests, previews, and hosted Agent work that must not touch production data.

From the Postgres service, create a named branch, capture its one-time connection URI, and delete the branch when the work is complete. Rotation replaces only that branch's credential. Hosted Agent sandboxes can request their own 24-hour branch through a scoped action and clean it up when the session ends.

A successful sandbox migration proves the migration against that branch. It does not migrate production. Production remains an explicit customer-owned CI step; see Run database migrations.

Manage connections in serverless code#

Reuse a small pool across warm invocations when your library supports it, but close or release work before the handler returns. Do not leave a blocking connection or worker loop alive solely to wait for future work. See Background workers.