Skip to content
LainstackDocs

Extract document text

Extracts page text, paragraph structure, word coordinates, and table cells from one document.

@lainstack/document/extract-text · Data

How to use it

Section titled “How to use it”

Extract document text reads one PDF, PNG or JPEG file and returns its text and layout as several dataset handles: pages, paragraphs, textItems, words, tableCells and issues. It reads layout and runs OCR. It does not understand content, so it returns no named fields such as totals or dates.

Typical pattern: extract the text, then pick the data you need.

  • Use @lainstack/ai/prompt to pull named fields out of page text.
  • Use @lainstack/flow/for-each-item to go through paragraphs or table cells.
  • Use @lainstack/collection/load-array to read a small collection as one list.

Traps:

  • The outputs are handles, not inline text. Reference a handle from a block that accepts datasets. Do not treat it as a string.
  • Table cells come as one row per cell, with table, row and column numbers. Rebuild rows yourself.
  • Coordinates are in points from the top-left corner of the page.
  • Limits: 50 MiB, 100 pages, 240 seconds, and image files up to 16 million pixels. A larger input fails the step. No partial result is published.
  • The file type is detected from the file content. A declared type that conflicts with it fails.
  • OCR language is English only.
  • issues is optional. It lists pages where paragraph, text item, word or table detail was not available.

Settings

Section titled “Settings”
Setting Key Type Details
File (required) file Value Extract text, paragraph structure, word coordinates, and table cells from one PDF or image.

Outputs

Section titled “Outputs”
Output Path Type
Source file sourceRef Value
Pages pages Value
Paragraphs paragraphs Value
Text items textItems Value
Words words Value
Table cells tableCells Value
Extraction issues issues Value
Page count summary.pageCount Number
Word count summary.wordCount Number
Table count summary.tableCount Number
Issue count summary.issueCount Number
Extraction engine summary.engine Text
Engine version summary.engineVersion Text
OCR language summary.ocrLanguage Text