PDF Text Extractor — fast PDF to text & JSON, per page

Give it PDF links; get back clean plain text — the whole document and every page separately — in true reading order. Multi-column pages are read column by column, not welded line by line across the gutter. Arabic and other right-to-left text keeps its direction. No OCR overhead, no API keys, no subscription.

Built for AI agents, RAG pipelines, and anyone who needs the text out of a PDF without ceremony.

What you get

One dataset item per input PDF:

{
  "url": "https://example.com/report.pdf",
  "status": "ok",
  "n_pages_in_document": 12,
  "n_pages": 12,
  "n_scanned_pages": 0,
  "script": "latin",
  "n_words": 4816,
  "n_chars": 31220,
  "text": "Full document text, pages joined with \f ...",
  "pages": [
    {
      "page": 1,
      "status": "ok",
      "scanned": false,
      "n_words": 402,
      "n_columns": 2,
      "text": "Page text in reading order ..."
    }
  ]
}

Turn on Include spatial layout rendering to also get text_layout per page — the page reproduced on a character grid (pdftotext -layout style), which preserves tables without ruling lines, forms and diagram labels.

What it does well

• Reading order on multi-column pages. Column gutters are detected from the page's whitespace geometry (up to 3 columns), full-width headings split the page into stripes, and each column is read top to bottom.
• Right-to-left text. Arabic lines are assembled right-to-left; Persian and Urdu ZWNJ orthography is preserved, not stripped.
• Honest scanned-page handling. A page with no usable digital text layer is flagged "scanned": true instead of silently returning empty text — and it is not billed. For scanned documents, use the sibling actor PDF OCR Extractor (Arabic + English OCR built in).
• Unicode hygiene. NFC normalization, zero-width and bidi control characters removed, exotic spaces unified — text that is safe to compare, index and embed.

Pricing — pay only for delivered text

Event	Price	Charged when
document-processed	$0.004	a PDF delivered at least one page of text
text-page	$0.0004	per page that actually carried text

A 10-page report costs $0.008. And the billing rule is deliberate: failed downloads, password-protected files, empty pages and scanned pages are never billed — not even the per-document fee. If you got no text, you pay nothing.

Your cost ceiling per document = $0.004 + $0.0004 × Max pages per document (default 100).

Input

Field	Default	Notes
pdf_urls	—	up to 100 direct PDF links per run
pages	all	'1-5', '1,3,7-9' supported
preserve_layout	false	adds text_layout per page
max_pages_per_document	100	hard cost/pages cap per PDF
include_failed_documents	true	failed docs appear with status + error

Limits, stated plainly

• Digital text layer only — no OCR in this actor. Scanned pages are detected and flagged; PDF OCR Extractor reads them.
• PDFs up to 50 MB, 100 URLs per run, up to 1000 pages per document.
• Password-protected PDFs are reported as errors (and not billed).

The family

Actor	Use it for
PDF Text Extractor (this one)	fast clean text from digital PDFs
PDF OCR Extractor	scanned/image PDFs, Arabic + English OCR, bounding boxes
PDF Table Extractor	tables as {headers, rows} JSON
Doc to Markdown	documents into LLM-ready Markdown

Issues and requests

Open an issue on the actor's Issues tab — I read all of them.