PDF to Text, Markdown & Tables – RAG Chunks with Page Citations

Convert any public PDF URL into clean, structured data in one call:

• 📝 Full text and per-page text
• 🧾 LLM-ready Markdown with headings, tables and page markers
• 📊 Table extraction: every table as row arrays and as a Markdown table, with its page number
• 🧠 RAG chunks with configurable size and overlap, each tagged with pageStart/pageEnd so your AI answers can cite the page
• 🔎 Find terms with page citations: give keywords like "total due", "termination" or "allergen" and get every match with its page number and surrounding quote
• 🏷️ Document metadata (title, author, creator, dates), page count, word count, SHA-256 hash and embedded links

Bulk-process hundreds of PDFs in parallel. No AI cost and no OCR fees. Failed downloads and scanned PDFs with no text layer are not charged.

Use cases

• RAG and AI agents: feed PDFs into vector databases (Pinecone, Qdrant, Weaviate, pgvector) with page-level citations
• Financial & legal docs: 10-Ks, annual reports, contracts, policies, filings. Pull the text and find clauses by keyword
• Invoices, quotes and spec sheets: extract line-item tables to CSV/Excel
• Research: papers, government reports, statistics releases (BLS, Census, FDA…)
• Restaurant & retail: PDF menus, price lists, catalogs

Input

{
  "pdfUrls": [{ "url": "https://www.bls.gov/news.release/pdf/cpi.pdf" }],
  "pages": "1-10",
  "outputMarkdown": true,
  "extractTables": true,
  "ragChunks": true,
  "chunkSize": 1500,
  "chunkOverlap": 150,
  "searchTerms": ["total", "allergen"]
}

Output (trimmed)

{
  "url": "https://www.bls.gov/news.release/pdf/cpi.pdf",
  "status": "ok",
  "pageCount": 38, "pagesExtracted": 38, "wordCount": 13442, "tableCount": 1,
  "metadata": { "Title": "Consumer Price Index - August 2026", "Producer": "Adobe PDF Library 26.2.141" },
  "markdown": "<!-- page 1 -->\n...\nCONSUMER PRICE INDEX – AUGUST 2026\n...",
  "tables": [{ "page": 6, "tableIndex": 0, "rows": [["..."]], "markdown": "| ... |" }],
  "chunks": [{ "chunkIndex": 0, "pageStart": 1, "pageEnd": 1, "text": "..." }],
  "searchHits": [{ "term": "total", "page": 5, "quote": "…The all urban consumer group represents over 90 percent of the total U.S. population…" }],
  "sha256": "…",
  "charged": true
}

Statuses: ok (charged), scanned_no_text_layer, fetch_error, access_denied, not_a_pdf, too_large, blocked_by_robots, parse_error (all free).

Test results (Oct 10, 2026)

PDF	Pages	Words	Tables	Result
IRS Form W-9	6	6,270	3	ok
BLS CPI news release	38	13,442	1	ok
"Attention Is All You Need" (arXiv)	15	1,746*	6	ok
SEC Form 10-K instructions	19	6,491	0	ok
FDA Food Code (668 pages, capped at 60)	60	19,095	4	ok
Restaurant PDF menu	1	627	0	ok
HTML page (not a PDF)	–	–	–	not_a_pdf (free)
Dead link	–	–	–	fetch_error (free)

*arXiv's PDF compresses spacing between words, so the word count runs low while the characters are complete.

Pricing

A flat price per PDF covers up to 50 pages. Longer documents add a small per-page charge for each page beyond 50 (see the Pricing tab). Use pages or maxPagesPerPdf to cap cost.

Limitations

• No OCR: scanned, image-only PDFs come back as scanned_no_text_layer and are not charged.
• Rotated margin text (for example vertical form labels) is dropped on purpose.
• Only public URLs. Login-walled or paywalled files are refused. robots.txt is respected.