Skip to main content
Welcome to the documentation for the Terrafai Document Extraction API. The Terrafai API allows you to extract structured data from unstructured documents such as invoices, receipts, bank statements, and other business documents. Using natural-language templates and intelligent document classification, the API converts documents into clean, machine-readable output that can be integrated directly into your systems. This documentation is intended for developers and technical teams who want to automate document processing using the Terrafai platform.

​
How the Terrafai API Works

At a high level, document extraction with Terrafai follows a simple flow:
  1. Define what data you want using templates
  2. Optionally classify documents using filters
  3. Submit documents for extraction
  4. Retrieve structured results
The sections below introduce the key concepts you’ll encounter throughout the guides.

​
Core Concepts

​
Templates

Templates define what information should be extracted from a document and how the output should be structured. A template consists of:
  • A set of fields written in natural language
  • Data types that control how values are parsed and formatted
  • Optional rules that transform or normalize extracted data
Templates are required for extraction unless you are using filters to automatically select a template.

​
Filters

Filters allow you to classify, route, or skip documents before extraction. They are designed for workflows where:
  • Multiple document types are submitted through the same integration
  • Different templates should be applied automatically
  • Irrelevant or unsupported documents should be ignored
When using filters, the system evaluates each document and decides:
  • Whether it should be extracted
  • Which template should be applied
This helps reduce manual routing and avoid unnecessary consumption.

​
Extraction Modes

The API supports two extraction modes:
  • Synchronous extraction
    Returns results in a single request–response cycle. Best for small documents and real-time use cases.
  • Asynchronous extraction
    Processes documents in the background and returns a job identifier. Best for large documents, batch processing, or long-running extractions.
Both modes use the same request structure and extraction options.