PDF to Markdown and JSON

This Actor takes a PDF that already has a text layer and gives you back Markdown and structured JSON, page by page. No OCR, no LLM calls. It reads the text that is already in the file and lays it out.

It's built for the case in which you have a report, an invoice or a paper, and you want its text in a format you can index or hand to a model.

What this is for

Reports, invoices, contracts, papers, anything born digital. If you can select the text in a PDF viewer, this works on it. If your PDF is a scan, meaning a picture of a page with no text layer, you get an empty page back. There is no OCR here.

Input

{
  "pdfUrl": "https://example.com/report.pdf"
}

or, to send the file directly instead of a URL:

{
  "pdfBase64": "JVBERi0xLjQKJ..."
}

Provide exactly one of pdfUrl or pdfBase64. PDFs up to 25 MB.

Output

One dataset item per run:

{
  "pageCount": 2,
  "markdown": "# Q3 Report\n\nRevenue grew...\n\n# Appendix\n\n| Region | Revenue |\n| --- | --- |\n| EU | $4.2M |",
  "pages": [
    {
      "pageNumber": 1,
      "markdown": "# Q3 Report\n\nRevenue grew...",
      "headings": [{ "level": 1, "text": "Q3 Report" }],
      "tables": []
    },
    {
      "pageNumber": 2,
      "markdown": "# Appendix\n\n| Region | Revenue |\n| --- | --- |\n| EU | $4.2M |",
      "headings": [{ "level": 1, "text": "Appendix" }],
      "tables": [
        {
          "rows": [
            ["Region", "Revenue"],
            ["EU", "$4.2M"]
          ]
        }
      ]
    }
  ]
}

Pricing

You pay per event:

Event	Price	When it's charged
pdf-processed	$0.01	Once per PDF that is successfully opened
page-extracted	$0.002	Per page extracted

So a 5-page PDF costs $0.02 and a 1-page PDF costs $0.012. Trying it out costs cents.

What it does well

Text extraction is the main job and what the pricing is built around. You get the prose of the document, per page and as one blob, which is what you want for RAG ingestion, search indexing, filling a context window, or simple PDF parsing.

Headings are detected from relative font size, so the Markdown keeps the document's structure instead of coming out as one flat wall of text.

Tables work, including borderless ones. That is the common case in reports, invoices and papers, where there are no ruled lines and the columns are just aligned text. Detection looks for column positions that repeat across a run of lines instead of requiring every row to look identical, so a row with a missing or extra cell still gets picked up, headers included. You get each table twice: as a Markdown pipe table inside the page text, and as JSON rows.

Limitations

• No images or figures. Only the text layer comes out.
• Math formulas come out as their plain characters. The text layer does not say which characters are a formula and which are prose.
• No OCR. Scanned or image-only PDFs return empty pages.
• Highly complex tables can degrade.
• 25 MB per PDF.
• Encrypted or password-protected PDFs are not supported.