PDF Text Extractor: PDF to Markdown & Text for LLMs

Turn PDFs into clean Markdown or plain text. Give it links to PDF files. For each document it returns the text layer as:

• Markdown: headings (detected from font size), paragraphs, bulleted and numbered lists, with page markers;
• or plain text: line by line.

Output is one item per document, or one item per page for chunking in RAG and LLM pipelines.

Try it free first

Run it with empty input. A free demo extracts the first 5 pages of RFC 9309, which isn't billed, so you can see the output format.

What it does and doesn't do

• Text layer only, no OCR. Scanned PDFs (images of text) have no text layer. Those pages come back empty, are listed in pagesWithoutText, and aren't billed.
• No tables or images. Table cells come out as text in reading order, and images are skipped.
• Headings are heuristic. They're detected from font size and spacing, so unusual layouts may get extra or missing headings.
• Untrusted files are handled safely. Scripts inside PDFs are never run and nothing is rendered. Each file is parsed in an isolated worker with time and memory limits.
• Languages: Chinese, Japanese and Korean PDFs are supported through pdf.js's bundled character maps. Exotic custom encodings may still lose text.

How to use it

1. Paste your PDF URLs (up to 100 per run, 50 MB each, 500 MB in total).
2. Tick I have the right to process these documents. Runs without it are refused.
3. Choose Markdown or plain text, and per-document or per-page output.
4. Optionally set Maximum pages per PDF (default 200, up to 500).

Output

Each dataset item has:

• url, title, totalPages, pagesRead, pagesWithText, pagesWithoutText, truncated, format and bytes;
• content for the document, or for each page with page.

Documents over 8 MB of text are saved whole to the run's key-value store, named by contentKey, and the item carries the first 200 KB. Failed documents carry an error:

• not_a_pdf, encrypted, invalid_pdf, parse_error, timeout or too_complex (the file needed too much memory);
• http_404 and other HTTP errors, too_large and too_many_redirects;
• blocked_by_robots, blocked_private_address, scheme_not_allowed or denied_host;
• duplicate_document (the same file under another URL), run_limit_reached or time_limit_reached.

Failed documents are never billed.

A SUMMARY record holds the counts.

Polite and safe by design

• robots.txt: honoured for every host, including Crawl-delay.
• Pacing: within a run, at most one request at a time per server, at least 500 ms apart, and at most 25 PDFs per site.
• Addresses: private and internal network addresses are refused at every redirect.
• What it never sends: cookies or credentials. It never tries to get around access controls or anti-bot protection.
• Identification: the user agent is PdfTextExtractor/1.0, with a link to this page.
• Site owners can opt out with User-agent: PdfTextExtractor and Disallow: / in robots.txt. To report misuse, open an issue on this Actor's Issues tab.

Pricing

• Pay per event: each page with real text (at least 20 characters) is one page-extracted event. At most 15 pages are billed per document, however long it is.
• Not billed: pages with fewer than 20 characters of text (blank or scanned pages, or a page number only), pages beyond your page limit, failed or duplicate documents, output that couldn't be saved, and the demo. A scanned page with a longer text stamp or banner counts as a page with text.
• The price is shown on the Pricing tab. You can cap the maximum charge per run in Apify's run options.

Terms of use

• Your rights: only process documents you own, that are licensed to you, or that you are otherwise allowed to use. You are responsible for copyright compliance.
• Personal data: if the documents contain personal data, you are responsible for having a lawful basis to process it.
• Where text goes: extracted text is stored only in your run's dataset and key-value store. The developer doesn't keep it or read it, and logs hold counts only.
• Limits per run: 100 PDFs, 50 MB per PDF, 500 MB in total, and 25 PDFs per site.

About

• Independent: made by an independent developer. Unofficial; not affiliated with or endorsed by Apify.
• Built with AI: built and maintained with the help of AI tools, under the developer's direction. Uses Mozilla's pdf.js (Apache-2.0).