PDF to Markdown, in your browser

How to do it

1. Work out which kind of PDF you have

   Try selecting a sentence in your PDF reader. If the text highlights word by word, it was made by software and already carries a text layer, which converts quickly and accurately. If your cursor draws a box over the whole page instead, it is a scan: a picture of a page, with no text in it at all. Both work here, but the second one has to be read by OCR, which takes longer and can misread.

2. Set the document language before you convert a scan

   OCR needs to know which alphabet it is looking at. English, Arabic, Spanish, French, German and Simplified Chinese are supported. This is the language of the document, not the language of this interface — an Arabic contract read with English selected will come back as nonsense that looks like words.

3. Choose fast or accurate

   Fast is the default and is right for a born-digital PDF. Accurate spends longer on each page and is worth it for a scan, a photograph of a page, or anything faint, skewed or low-resolution.

4. Check the tables and the order before you use it

   Open the review panel on the file's row. It reports whether anything was measured as uncertain, and says plainly when no position was recorded for a passage rather than guessing at one. A page that produced no text at all is reported as a partial result rather than passed off as an empty page.

A worked example

A two-column report page with a running header on every page. The header is in the top eighth of each page and repeats, so the optional cleanup can offer to remove it; the body text is left alone even where it says the same words.

[page 4 of 20]
  ACME Quarterly Report            <- repeated on every page
  Results
  Revenue rose in the second
  quarter, driven by the night
  freight route.
                      - 4 -        <- page number in the margin

## Results

Revenue rose in the second quarter, driven by the night freight route.

The converter has two PDF samples you can try without a file of your own: a table with a real text layer, and a scanned page with no text layer at all.

What it will not do

• A scan is read by OCR, and OCR makes mistakes. The result is a reading of the page, not a certified transcript, and a confidence figure is not a percentage of correctness.
• A PDF with no text layer and no readable image — a blank scan, or one too faint to read — produces a partial result. It is reported as partial rather than returned as an empty page.
• Multi-column layouts can be read in the wrong order. Where the reading order was measured as uncertain, the review panel says so.
• A table drawn with lines and spacing rather than as a real table may come out as paragraphs. Check any table you rely on.
• A password-protected PDF needs its password. It is used in this tab and never stored or sent.
• Form fields, annotations, digital signatures and embedded attachments are not carried into the Markdown.

Questions

Does my PDF get uploaded anywhere?

No. It is read by code running in this tab. The file, its name and its text never leave your device.

Why is the scanned page taking so long?

The first scan you convert downloads an OCR model for the language you picked. That happens once; afterwards the same language works offline. Accurate mode also deliberately spends longer per page than fast mode.

Can it handle a PDF in two languages?

You can select more than one document language, and OCR will look for all of them. More languages means more work per page, so pick only the ones actually in the document.

How many tokens will the result be?

The review panel counts the Markdown you are about to copy or download, and names the method it used. Today that method is an approximation that runs on your device, not a tokenizer, so treat it as a guide rather than a bill. Real counts differ by model and by language.

Is there a page limit?

There is a limit that depends on your device, because the work happens on it. A long PDF on a phone will refuse earlier than the same file on a laptop.

Where your file goes

Nowhere. Reading, converting, cleaning and counting all happen in this browser tab, on your own device. The file, its name and its text are never uploaded to a server, and there is no account to create.

One thing does use the network, and it is worth being precise about: reading a scanned page needs an OCR model, which downloads from this site the first time you convert an image or a scan. That is a program arriving on your device, not your document leaving it. After it has been fetched once, that conversion works offline.