Document extraction guide

This guide documents the extraction API with both supported modes: vision and layout.

Guide structure

1. Modes
2. Extract schema
3. Response structure

Modes at a glance

Mode	Best for	Notes
vision	Handwritten forms, free-layout scans, image-heavy pages	Vision-language extraction mode
layout	Stable business documents such as invoices, orders, and contracts	Structured extraction mode with optional grounding

Recommended reading order

• Start with Modes to understand the differences and selection advice for the two modes
• Then read Extract schema to learn how to write extract schema
• Finally read Response structure to understand how the returned results carry field values and grounding information