PDF, Image & Office to Text and Markdown (with OCR)

Turn any document into clean text and LLM-ready Markdown in one Actor: PDFs (digital or scanned), images, Word (.docx), PowerPoint (.pptx) and Excel (.xlsx). Tables become Markdown tables and structured rows, scanned pages are read with OCR in 24 languages, and you can get RAG-ready chunks directly.

Why this Actor

• One Actor for every format: PDF, PNG, JPG, TIFF (multi-page), WebP, BMP, GIF, DOCX, PPTX, XLSX. No need to chain a PDF parser, an OCR tool and an Office converter.
• Smart OCR, cheaper runs: in Auto mode the PDF text layer is used when it exists and OCR runs only on scanned pages. You pay the OCR price only for pages that need it.
• Markdown with structure: headings and lists from Word, slide titles and speaker notes from PowerPoint, one table per Excel sheet, and real tables from PDFs placed where they appear on the page.
• RAG-ready: set a chunk size and get the Markdown split on paragraph boundaries.
• Google Drive and Dropbox links work as they are (share links are converted to direct downloads).
• No start fee: you pay per page actually processed. A broken or missing file is not charged and does not stop the run.

Use cases

• RAG and AI agents: feed PDFs, decks and spreadsheets to your LLM as Markdown chunks
• Document processing: invoices, contracts, reports, forms and scans to searchable text
• Data extraction: tables from PDF reports and Excel files as rows of cells
• Archives: OCR of scanned books, letters and faxes in 24 languages

Input example

{
  "documents": [
    "https://arxiv.org/pdf/1706.03762",
    "https://example.com/scanned-invoice.pdf",
    "https://drive.google.com/file/d/FILE_ID/view",
    "https://example.com/quarterly-report.docx"
  ],
  "ocrMode": "auto",
  "ocrLanguages": ["eng", "deu"],
  "chunkSize": 2000
}

Output example

One item per document:

{
  "url": "https://example.com/quarterly-report.docx",
  "fileName": "quarterly-report.docx",
  "fileType": "docx",
  "pageCount": 1,
  "pagesProcessed": 1,
  "ocrPages": 0,
  "truncated": false,
  "text": "Quarterly report\nRevenue grew 12% year over year.\n…",
  "markdown": "# Quarterly report\n\nRevenue grew 12% year over year.\n\n| Quarter | Revenue |\n|---|---|\n| Q1 | 100 |\n| Q2 | 112 |",
  "wordCount": 18,
  "tableCount": 1,
  "chunks": ["# Quarterly report\n\nRevenue grew 12%…"],
  "pages": [
    { "page": 1, "text": "…", "markdown": "…", "tables": [[["Quarter", "Revenue"], ["Q1", "100"], ["Q2", "112"]]], "ocr": false }
  ],
  "error": null
}

What a "page" is: a PDF page, an image (or a TIFF frame), a PowerPoint slide, an Excel sheet; a Word document counts as one page.

Options

• OCR: Auto (only scanned pages), Always (every PDF page, e.g. PDFs with a broken text layer), Never (text layer only)
• OCR languages: English, German, French, Italian, Spanish, Portuguese, Dutch, Polish, Swedish, Danish, Norwegian, Finnish, Czech, Romanian, Hungarian, Turkish, Russian, Ukrainian, Greek, Chinese (simplified), Japanese, Korean, Arabic, Hindi
• Pages (e.g. 1-5,8) and max pages per document
• Per-page content, tables as data, RAG chunk size, OCR resolution

Pricing

Pay per page: a low price for pages with a text layer and for Office documents, a higher one for pages that need OCR (see the Pricing tab). Set a maximum charge per run in the run options: the Actor stops cleanly when it is reached and marks the cut documents with truncated: true.

Tips

• Files must be reachable by a public link. To use a file from your computer, upload it to an Apify key-value store, Google Drive or Dropbox and paste the link.
• For scans, choose only the languages that are in the document: OCR is faster and more accurate.
• Password-protected PDFs cannot be read.

Privacy

Documents are downloaded, processed in memory during the run and returned only to your dataset: they are not stored or shared elsewhere. At the end of each run the developer receives an anonymous notification with only numbers (pages processed, amount charged, duration, success or failure). Your input, the documents and your identity are never sent.