Document → Structured Data Extractor

PDF or image documents in — structured data out. Feed it document URLs; get back the full text, detected tables as rows you can export to CSV, and pattern-detected business fields — dates, amounts, totals, reference/invoice numbers, emails, phones — as clean JSON.

Most PDF extractors dump raw text and stop. This one gives your workflow something it can actually use downstream: rows, fields, and per-page provenance.

What you get per document

Output	Description
text	Full extracted text (per-page, layout-aware line reconstruction)
tables[]	Detected tables as rows[][] — export straight to CSV/Excel
fields{}	dates, amounts, totals, referenceNumbers, emails, phones
pages[]	Per-page character counts + extraction method (text-layer vs OCR)
kind, pagesProcessed, error	Type detection and honest failure reporting

Pricing — pay only for work actually performed

Event	Price
Document processed (tables + fields extraction)	$0.01
Text-layer page	$0.001
OCR'd page (images / scans)	$0.02

Examples: a 20-page digital PDF ≈ $0.03. A 200-page digital PDF ≈ $0.21. 50 scanned images ≈ $1.01. Failed downloads and unparseable documents are not charged.

What it does NOT do (honesty section)

• No LLM, no AI guessing — everything is deterministically parsed or OCR'd; fields are regex/pattern-detected and should be spot-checked for critical use.
• Scanned PDF pages (image-only, no text layer) are flagged needsOcr: true and not charged in this version — direct image files (PNG/JPG/TIFF) are OCR'd. Send scans as images for OCR.
• No login-protected documents, no DRM circumvention.
• Table detection is conservative — it prefers missing an ambiguous table over fabricating rows.

Input

{
  "documents": ["https://example.com/invoice.pdf", "https://example.com/scan.png"],
  "ocr": true,
  "language": "eng",
  "maxPages": 50
}

Up to 200 documents per run, 500 pages per document, 25 MB per file. OCR supports Tesseract language codes (eng, deu, fra, spa, …).

Typical uses

• Invoices / receipts → amounts, dates, reference numbers for bookkeeping automation
• Product catalogs / spec sheets → tables to CSV
• Reports and filings → searchable text + key figures
• Contact-bearing documents → emails and phones with document provenance

FAQ

Why not a flat per-page price? Because a text-layer page costs ~40× less to process than an OCR page. You shouldn't subsidize other people's scans.

What about scanned multi-page PDFs? v1 flags those pages (needsOcr) without charging. Convert to images for OCR, or watch for the next version.

Is my document stored? Documents are fetched, processed in the run's container, and the extracted results are written to your dataset. The Actor keeps no copy outside your run's storage.