Overview

The API extracts structured data from unstructured documents like PDFs and images. It identifies elements such as text, tables, and form fields, and returns them in hierarchical JSON with exact page and coordinate references.

understands relationships between elements, like captions linked to images, and works without templates or training.

The extracted data can be used in downstream applications such as retrieval-augmented generation (RAG), search, or other custom workflows.

​

Try Out Agentic Document Extraction

​

Features

• Layout-agnostic parsing: Extracts data from complex layouts. No training or templates needed.
• Element detection: Identifies specific elements including text, tables, form fields, checkboxes, and more.
• Understands hierarchical relationships: Detects how elements relate in structure and meaning. For example, can understand that a line of text is the caption for an image.
• Precision extraction: Extracts data accurately, even from complex documents.
• Flexible output: Returns results in Markdown and JSON, ready for use in downstream applications like retrieval-augmented generation (RAG).
• Visual grounding: The JSON output includes the document, page, and coordinate-level references for each element to support traceability, validation, and compliance workflows.
• Supports multiple file types: Can extract data from PDFs and common image formats.

​

Introduction from Andrew Ng