Soft delete

Mark a date column .softDeleteAt(). Reads then automatically skip rows where that column is set, and you get a dedicated verb to set it, an inverse to clear it, and the _withDeleted escape hatch to read past the filter.

delete() / deleteMany() are always hard deletes — they remove the row, exactly like Prisma, regardless of whether the model has a soft-delete column. The recoverable path is explicit:

const Account = model('accounts', { id: f.id(), deleted_at: f.dateTime().softDeleteAt() });

// Soft delete — sets deleted_at; the row stays but is hidden from reads.
await db.account.softDelete({ where: { id: 'a1' } });
await db.account.softDeleteMany({ where: { tenantId: 't9' } });

// Reads skip soft-deleted rows by default; opt back in with _withDeleted.
await db.account.findMany();                                   // excludes a1
await db.account.findMany({ where: { _withDeleted: true } });  // includes a1

// Restore — clears deleted_at, row is active again.
await db.account.restore({ where: { id: 'a1' } });
await db.account.restoreMany({ where: { tenantId: 't9' } });

// Hard delete — permanently removes the row (and runs cascades). No going back.
await db.account.delete({ where: { id: 'a1' } });
await db.account.deleteMany({ where: { tenantId: 't9' } });

softDelete, softDeleteMany, restore, and restoreMany throw if the model has no .softDeleteAt() column — use delete for a hard delete in that case.

Upgrading from v1.x? In v1, delete() / deleteMany() silently soft-deleted on models with a .softDeleteAt() column. v2 makes delete() always hard. Search your codebase for delete( / deleteMany( calls on soft-delete models and switch the ones that expected recoverable behavior to softDelete() / softDeleteMany(). See the CHANGELOG for the full migration note. This is a runtime semantic change — it will not show up as a type error.

See more — docs/SOFT-DELETE.md for partial-filter uniques that ignore soft-deleted rows, cascade restore semantics, retention-purge cron, FTS / vector / search-index interaction, GDPR caveats (soft-delete is not erasure), and 3 worked patterns.

---