Upstage Information Extract icon

Upstage Information Extract

Extract structured data from documents/images using Upstage Information Extraction

Actions2

Extract Information

AI-generated

Summary

Extract structured data from a document or image using Upstage Information Extraction.

Inputs

  • inputType — Method of providing the document or image. Binary reads from a previous node's binary property; URL uses an image URL string. Default: binary.
  • model — Upstage model for information extraction. Only option is 'information-extract' (recommended).
  • returnMode — Determines output format. 'extracted' returns parsed JSON + model + usage + full_response; 'schema' returns schema_type, json_schema, raw, model, usage; 'full' returns complete API response object. Default: extracted.
  • inputType — When inputType is binary, binaryPropertyName specifies the property name containing the file (default: 'document'). When inputType is url, imageUrl provides the image URL string (must use http:// or https://).
  • schemaInputType — How to provide the JSON schema for the extraction. 'schema' uses a dedicated schema object with schemaName and json_schema; 'full' uses a complete response_format JSON object. Default: schema.
  • schemaName — Name for the JSON schema in the response_format (default: 'document_schema'). Visible when operation is 'extract' and schemaInputType is 'schema'.
  • json_schema — Target JSON schema for extraction defined as an object schema (default: { type: 'object', properties: {} }). Visible when operation is 'extract' and schemaInputType is 'schema'.
  • fullResponseFormat — Complete response_format JSON including type, json_schema, name, and schema (default provided). Visible when operation is 'extract' and schemaInputType is 'full'.
  • pagesPerChunk — Number of pages per chunk for processing (recommended for documents with 30+ pages). Set to 0 to disable. Visible when operation is 'extract'. Default: 0.
  • location — When true, includes page number and coordinates for each extracted value. Visible when operation is 'extract'. Default: false.
  • confidence — When true, includes confidence score (high/low) for each extracted field. Visible when operation is 'extract'. Default: false.
  • split — When true, processes multi-page documents and returns split results per page. Visible when operation is 'extract'. Default: false.

Output shape

The returned data structure depends on returnMode:

All outputs include the API response model and usage metadata. Binary data from the input item is not passed through during the extract operation (only during schema generation).

Links

Discussion