Database

A Database is one customer Postgres. It is not a schema on a shared server: a database has its own engine, its own credentials, its own backup, its own region, and its own deletion. Nothing you do to yours can be seen from anyone else's.

#What you choose, and what you do not

You choose the Postgres major version, the compute size, the region, and whether deletion is protected. You do not choose an instance class, a disk size, a replica count or max_connections: those derive from the compute size, so a database cannot be configured into a state it cannot serve.

An idle free-tier database stops, and the next connection wakes it. That is a state you do not manage: nothing to start, nothing to shut down.

#Every database is addressable

A database's name is a path, and the same path works in the API, the CLI and the console:

Name

orgs/{org}/projects/{project}/envs/{env}/databases/{database}

Id

db_<location><ulid>

Read a database, list the ones in an environment, create, update, delete, open a connection, rotate its credentials and restore it in place — the databases collection has every call, with its scope and its examples. data:read covers the reads; everything that changes the database needs data:write.

#Postgres, with the extensions you would install yourself

pgvector is part of the engine, so a vector column is a normal column:

CREATE EXTENSION IF NOT EXISTS vector;

CREATE TABLE documents (
  id        uuid PRIMARY KEY DEFAULT gen_random_uuid(),
  content   text NOT NULL,
  embedding vector(1536)
);

CREATE INDEX ON documents USING hnsw (embedding vector_cosine_ops);

#Backups and point-in-time restore

A database is backed up on its own schedule, and restore puts it back to a point in time inside its retention window. The restore happens in place: the database is unavailable while it restores, and everything written after the restore point is gone.

#Connections are issued, not constructed

There is no connection string in the console to copy into a repository: you ask for one with connect, and it is returned to that caller only. The response carries the pooled uri and — for Postgres — the direct_uri that migrations, pg_dump, LISTEN/NOTIFY, session SET and logical replication need.