Document to Markdown, OCR & Invoice Extraction

Turn PDFs, Word, PowerPoint, Excel and HTML files into clean Markdown for LLMs and RAG, read text from images and scanned PDFs, or pull invoice, receipt and form data into JSON that matches your own schema. Paste file URLs, pick a mode, get one dataset row per file.

What it does

Mode	Input	Output
Document to Markdown	PDF, DOCX, PPTX, XLSX, XLS, HTML (up to 50 pages)	Markdown with headings, lists and tables kept. Scanned PDF pages are OCR'd automatically. Optionally the full document structure as JSON.
OCR	PNG, JPG, TIFF, BMP, GIF, WebP, PDF	Plain text, overall and per page (English)
Extract data to JSON	Image or PDF up to 5 pages, plus your JSON Schema	JSON that fills your schema: vendor, totals, dates, line items, names, anything you define

Typical speed: Office and HTML files ~0.4 s, PDFs ~4 s per page, OCR ~0.5 s per image, extraction ~4 s per document. Files up to 20 MB.

Why use it

• Built for LLM pipelines: Markdown keeps document structure (headings, tables), so chunking and retrieval work better than with raw text.
• No templates for extraction: describe the fields you want as a JSON Schema and the vision model fills them, for any invoice or receipt layout.
• Private: files are processed in memory on dedicated servers and deleted as soon as the job ends. Nothing is stored or used for training.
• Pay only for results: failed files are never charged.

Input example

{
  "mode": "extract",
  "fileUrls": ["https://example.com/invoices/inv-1042.pdf"],
  "schema": {
    "type": "object",
    "properties": {
      "vendor": {"type": "string"},
      "invoice_number": {"type": "string"},
      "date": {"type": "string", "description": "YYYY-MM-DD"},
      "total": {"type": "number"},
      "line_items": {"type": "array", "items": {"type": "object", "properties": {
        "description": {"type": "string"}, "amount": {"type": "number"}}}}
    }
  }
}

Output example

Markdown mode:

{
  "url": "https://example.com/report.pdf",
  "fileName": "report.pdf",
  "status": "ok",
  "pages": 5,
  "format": "pdf",
  "markdown": "## Operations Report\n\n| Region | Revenue |\n|---|---|\n| West | $1.2M |",
  "seconds": 14.5
}

Extract mode:

{
  "url": "https://example.com/receipt.png",
  "fileName": "receipt.png",
  "status": "ok",
  "pages": 1,
  "data": {"vendor": "Bluefield Logistics", "total": 285.94},
  "seconds": 2.1
}

A file that can't be processed gets "status": "failed" and an error explaining why (for example an unsupported file type or a download error), and isn't charged.

Pricing

Pay per result:

• Document to Markdown: per page (a Word, PowerPoint, Excel or HTML file counts as one page)
• OCR: per page
• Extract: per document

See the Pricing tab for current prices. Set a maximum cost per run in the run options and the Actor stops before going over it.

Tips

• URLs must be reachable without logging in (public links or signed URLs from S3, Google Cloud Storage, Dropbox and similar).
• The file type comes from the URL's extension, or the server's Content-Type if there isn't one.
• For extraction, descriptive field names and descriptions ("YYYY-MM-DD", "amount without currency symbol") improve results.
• Use it from your own code through the Apify API, or chain it after a crawler that collects document links.

Limits

20 MB per file. Markdown: 50 pages per PDF. Extraction: images or PDFs of up to 5 pages. OCR is English. Results can contain errors: check extracted values before relying on them for financial or legal decisions.