Adapters

The read shape both adapters share

A read returns the record's stored fields plus id. An adapter neither invents fields nor hides them, so the same record reads back the same way whichever backend holds it:

{ id: "hero", title: "Hello", createdAt: "2026-09-06T…", updatedAt: "…" }

• id is the record's address, and so is the collection name. Neither is a field, and collection is not returned.
• createdAt / updatedAt are the adapter's own, written on every create and update, and both adapters return them.
• Timestamps are ISO strings, not Date objects, on both backends. These records are JSON-serialized across createContentHandler and server-rendered payloads, where a Date becomes a string anyway.
• An adapter accepts back what it returned: handing a record you just read to createWithId or upsert works, which is what makes copying one environment into another possible.

seedItemMap strips all four before writing, since they address the record or belong to the adapter rather than being content.

better-content/adapters/postgres

class PostgresDataAdapter implements DataAdapter {
  constructor(config: PostgresAdapterConfig);
}

interface PostgresAdapterConfig {
  db?: PgDatabase;                     // a Drizzle database (any pg driver)
  pool?: Pool;                         // or a pg Pool to build one from
  connectionString?: string;           // or a connection string
  schema: Record<string, PgTable>;     // collection name → Drizzle table (required)
}

Typed-only Drizzle adapter. You declare tables with pgTable(...) and own migrations (Drizzle Kit); the adapter performs DML only.

• Throws on unregistered collections, undeclared fields (writes and filters), and unsupported query shapes.
• contains maps to ILIKE '%value%'; in/nin map to inArray/notInArray; OR groups are supported.
• Default ordering: createdAt descending when the column exists and no orderBy is given.
• update/upsert set updatedAt = new Date().
• Reads return createdAt/updatedAt as ISO strings, and writes accept either those strings or a Date.
• create generates an id with crypto.randomUUID where available.
• pg and the node-postgres driver load lazily, only when the adapter must build its own pool; passing db works without pg installed, including in browsers (PGlite).

Peers: drizzle-orm >= 0.40, pg >= 8 (pool path only).

better-content/adapters/firestore

class FirestoreDataAdapter implements DataAdapter {
  constructor(config?: FirestoreAdapterConfig);
}

interface FirestoreAdapterConfig {
  db?: Firestore;                      // existing admin Firestore instance
  credentials?: {                      // or service-account credentials
    projectId?: string;
    clientEmail?: string;
    privateKey?: string;
    databaseURL?: string;
  };
  defaultOrderByField?: string;        // default "createdAt"
}

Operator mapping: eq ==, ne !=, lt <, lte <=, gt >, gte >=, in in, nin not-in.

• Throws on contains (no native substring search) and on OR filter groups; the errors say so explicitly.
• Firestore Timestamps serialize to ISO strings in results.
• create/createWithId stamp createdAt and updatedAt; update/upsert refresh updatedAt (upsert merges).
• createWithId uses Firestore's create, so it rejects an id that already exists rather than overwriting it. Use upsert to write regardless, or delete then createWithId to replace a document.
• Because reads order by createdAt and Firestore omits documents that lack the ordering field, records written only through upsert can be missing from an unfiltered read. createWithId stamps createdAt, so seeding through it (or through delete + createWithId) avoids this.
• With no query, reads order by defaultOrderByField descending.

Peer: firebase-admin >= 12 (uses the modular firebase-admin/app and firebase-admin/firestore APIs).