Schema and migrations

Generate the canonical DDL for your naming configuration:

import { createPostgresMemorySchemaSql } from '@anvia/memory-postgres'

const sql = createPostgresMemorySchemaSql({
  tablePrefix: 'assistant_',
})

The SQL creates pgcrypto, session, message, and error tables plus the unique message-position index. Apply it with your migration system, review privileges, and call store.validate() at runtime.

The session table includes nullable JSONB compaction_state. It stores the latest summary checkpoint and canonical message boundary; compaction does not remove covered messages. The generated SQL uses ADD COLUMN IF NOT EXISTS for an existing adapter-managed table. Application-managed deployments should apply and verify that additive migration before the new adapter version starts.

The schema uses generated UUIDs and cascading foreign keys. Renaming a prefix or explicit table name points the adapter at a different schema; it does not rename existing objects or copy data.

Re-generate and diff DDL after package upgrades. The source changelog records contract changes, but your deployment owns rollout and rollback.