Casing

Set casing in the config. A config without it uses 'snake', which keeps database names as they are; better-supabase init writes casing: "camel" unless you pass --casing snake.

export default defineConfig({ casing: "camel" });

With 'camel', models, filters, includes and returned rows use camelCase. The mapping happens in the query itself, not in a post-processing pass:

await db.customers.findMany({ select: ["id", "organizationId"] });
// GET /customers?select=id,organizationId:organization_id

PostgREST renames the columns in its response, so rows need no transform and nested includes keep their shape.

Per-table overridesCopy Anchor Link

export default defineConfig({
  casing: "camel",
  tables: { legacy_events: { casing: "snake" } },
});

Function resultsCopy Anchor Link

db.$rpc returns rows of a table (returns setof customers) and records (returns table (customer_id uuid, total numeric)) in the configured casing, with the same codecs as reads. gen records which functions need it in the metadata, so other results, such as a scalar, json or jsonb, come back as Postgres sent them:

const rows = await db
  .$rpc("customers_by_status", { p_status: "lead" })
  .orThrow();
rows[0].organizationId;

returns validates the decoded value, so a schema for the cased row works. { raw: true } skips the decoding and returns database names, typed as unknown. PostgREST sends int8 as a JSON number in function results, so values above 2^53 lose precision before a bigint codec sees them; cast them to text in the function when they can be that large.

What stays in database namesCopy Anchor Link

• Constraint names (onConflict: 'customers_organization_id_kvk_key').
• Plugin config columns (softDelete: { column: 'archived_at' }): these name database columns, and the generated flags hold the app names.
• Raw SQL, RPC argument names and the keys inside json results.

Error details keep the database name in constraint and columns, since they come from Postgres.