Structured record

A persistent entity represented through named fields, types, identifiers, and relationships under a defined schema.

A structured record is a fact stored in labeled fields rather than loose prose, so an assistant can filter, sort, and cross-reference it directly. Each piece of information sits in its own named slot, which lets you ask precise questions of the whole collection. Picture a contact saved as four fields: name, role, last contacted, and open topics. With that shape, you can tell the assistant to list everyone you have not contacted in thirty days who still has an open topic, and it answers by reading the fields. Stored as a paragraph instead, the same facts would force the assistant to reread and guess at every record.

Builder example

Typed fields support deterministic filters and validation, while free text preserves nuance the schema may omit. Generated extraction can still choose the wrong entity, date, or status, so provenance, constraints, and exception handling matter. Schemas should evolve through migrations rather than pretending the first field list is final.

Common confusion: A structured record organizes the fields you store; structured output constrains the fields and format the model returns in one reply. You can ask for structured output and then file it as a structured record, and the record persists across sessions while a single reply does not.