Storage

The Storage page manages two distinct capabilities:

• Collections are durable, typed PostgreSQL data shared by explicitly granted Functions.
• Cache is non-authoritative Redis data with bounded operational inspection.

The existing ctx.storage key/value API remains available for simple Function-private state.

Typed collections

A collection is a Project resource with immutable JSON Schema versions and declared PostgreSQL indexes. Records are separated by environment; there is no tenant partition inside a collection. Schema versions and Function grants affect runtime traffic only after a Project deployment; production receives them through the normal immutable release process.

Functions access granted collections through ctx.collections:

const customers = ctx.collections.collection<{
  name: string;
  status?: string;
  score?: number;
}>("customers");

const page = await customers.query({
  where: {
    and: [
      { field: "status", op: "eq", value: "active" },
      { field: "score", op: "gte", value: 10 },
    ],
  },
  orderBy: [{ field: "score", direction: "desc" }],
  limit: 50,
});

Available operations are create, get, query, count, update, and delete. Updates and deletes require the current record revision. Filtering, ordering, counting, limits, and cursor predicates execute in PostgreSQL; records are never fetched wholesale for application-side filtering.

Collections support explicit read, write, and delete Function grants. Calls without the needed grant, invalid schemas, and stale revisions fail safely.

Declared B-tree indexes support scalar/composite filtering and sorting. GIN indexes support JSON or array containment. Unindexed queries remain bounded and database-side but may scan, so the Storage page displays declared index coverage for review.

Record inspector

Owners and admins can select an environment, run the same bounded query DSL, create or edit schema-validated JSON records, and permanently delete records with revision checks. Developers can manage collection schemas but cannot inspect record values. Every mutation creates an immutable audit event.

Cache inspector

Cache inspection uses Redis SCAN, not KEYS. Owners and admins can filter by tenant or key prefix and inspect key scope, TTL, and serialized size. Revealing a value is an audited operation and is limited to 256 KiB. Known platform secret values, bearer tokens, and secret-like fields are always masked. Individual keys may be deleted with an audited operation.