PDF to Text Extractor: batch, pay only for successful PDFs

Extract clean per-page text and document metadata from text-based PDFs by URL, up to 50 per run. $0.003 per successfully extracted PDF, no start fee, no OCR. Built for automation pipelines (RAG ingestion, document indexing, search, data entry).

Use cases

• Ingesting public PDFs (forms, papers, reports) into a RAG or search index
• Document indexing and data-entry pipelines
• Extracting the text layer before table or layout processing
• Bulk extraction from a list of PDF URLs, paying only for the PDFs that worked

What you get

For each PDF:

• status: ok, no_text_layer, encrypted, invalid_pdf, download_failed, extraction_failed, resource_limit, or error
• text: text from the requested pages, up to your configured page limit
• pages[]: text per page, with character counts (optional)
• pageCount, pagesExtracted, truncated
• metadata: title, author, creator, producer, creation and modification dates
• processingMs, and error with a plain-language reason when something fails

Pricing: you only pay for PDFs that worked

One event per successfully extracted PDF ($0.003). Downloads that fail, scanned (image-only) PDFs, password-protected PDFs and corrupt files are reported in the output but never charged. There is no start fee. Page extraction errors and resource-limit failures are also uncharged. The run checks your spending limit before each PDF and stops cleanly when it's reached. A successful extraction of your configured page range is charged once even when truncated is true; pages beyond that range are not processed.

Accuracy

The output is tested against an independent reference extractor (Poppler pdftotext) on real public documents (IRS forms, arXiv papers, a NIST publication): word recall 97.9–100%, word precision 96.5–100%. Reading order can differ on multi-column layouts. The test suite also checks exact text on authored documents, including numbers, currency and accented characters.

Limits (by design)

• No OCR. PDFs with no extractable non-whitespace text in the requested pages return no_text_layer. Mixed image/text documents return only their existing text layer; images are not transcribed.
• Max 50 URLs per run, max 50 MB per file (default 25), and max 500 pages per PDF (default 200).
• Only public http(s) links; local and private-network addresses are refused.
• Each PDF has a 25-second processing deadline and a 500,000-character text limit. Hosted parsing runs in a separate process with a 384 MiB memory limit. Files exceeding these limits return resource_limit without an extraction charge.
• No table structure detection: tables come through as text.

FAQ

Does it work on scanned PDFs? No — there is no OCR. PDFs with no text layer return no_text_layer and are not charged. Use an OCR actor for scans.

Am I charged for failed downloads or encrypted PDFs? No. Only successful extractions are charged; every other outcome is reported with a status and not charged.

Can I extract only certain pages? Yes. You configure the page range/limit per run; pages beyond it are not processed, and a truncated-but-successful extraction is charged once.

How are tables handled? There is no table structure detection; tables come through as text.

What are the limits? Up to 50 public http(s) URLs per run, 50 MB per file (default 25), 500 pages per PDF (default 200), a 25-second processing deadline and a 500,000-character text limit per PDF.

What does it cost? $0.003 per successfully extracted PDF, no start fee.

Input example

{
  "pdfUrls": ["https://www.irs.gov/pub/irs-pdf/fw9.pdf"],
  "maxFileSizeMb": 25,
  "maxPagesPerPdf": 200,
  "includePageText": true
}