Paperly / PDF to Markdown
How to convert a PDF to Markdown without uploading it
Most PDF-to-Markdown tools ask you to upload the file, wait for a server, then download a result. Paperly does the work inside the browser tab instead: the PDF is parsed by pdf.js on your own machine, and the Markdown is built there too.
Convert a PDF now ↗Why local conversion matters
A PDF is rarely a neutral document. Contracts, medical records, financial statements, internal specs and unpublished drafts all travel as PDFs, and "we delete it after two hours" is a promise, not a guarantee. When the conversion runs in the tab, there is nothing to delete because nothing left the machine.
That has a practical side effect: conversion keeps working on a plane, in a locked-down corporate network, or behind a proxy that blocks file uploads. The only thing that ever downloads is the OCR model, and only if a page turns out to be a scan.
The three realistic options
| Approach | Good at | Trade-off |
|---|---|---|
| Online converter | Speed, no install | File leaves your machine; per-page limits and paywalls |
| CLI tools (Pandoc, Marker, MinerU) | Batch jobs, scripted pipelines, heavier models | Python environment, model downloads, GPU for the good ones |
| Browser converter | Privacy, zero setup, works offline after first load | Geometric heuristics instead of a vision model, so complex layouts need review |
Step by step
- Open the converter. Nothing to install, no account.
- Drop the PDF in. The file is read with the File API and stays in memory.
- Press Convert. Text-layer PDFs finish in seconds. If a page has no extractable text, it is rendered to a canvas and read by OCR.
- Check the structure. Headings, list items and simple tables are rebuilt from font size, weight, baseline spacing and column geometry.
- Copy or download. Copy as rich text for Notion or Docs, or download
.mdfiles, optionally in a ZIP with YAML frontmatter for Obsidian.
What the converter actually reconstructs
- Headings from font size relative to the body text, with weight as a tie-breaker.
- Lists from bullet glyphs and consistent indentation.
- Simple tables from aligned column positions on consecutive lines.
- Paragraphs from baseline gaps, so wrapped lines join and real paragraph breaks stay.
- Running headers and footers are detected across pages and stripped.
Two-column layouts are read in the correct order by clustering text blocks into columns before sorting them.
Scanned PDFs
If a page contains fewer than roughly thirty characters of real text, it is treated as a scan. The page is rendered at double resolution and passed to a Tesseract worker compiled to WebAssembly. English and Simplified Chinese models are available, and the model file downloads once, then stays cached.
Accuracy on a clean 300 DPI scan is good. Skewed photos, stamps over text and low-contrast faxes are not, and OCR of tables is the hardest case. More detail on that in OCR for scanned PDFs.
After the conversion
Markdown is only useful once it lands somewhere. The tool can export the same content back out as HTML, Word, Excel, PowerPoint, EPUB or an image, and the browser's print dialog produces a PDF. These exports are deliberately basic: Word export may drop list numbering and links, Excel puts each Markdown table on its own sheet, PowerPoint uses top-level headings as slide breaks.
Questions
Does converting a PDF to Markdown upload my file?
No. The PDF is parsed inside the tab with pdf.js. There is no server endpoint that accepts documents, which is also why the page keeps working after you go offline.
Will tables survive the conversion?
Simple tables with clean column geometry become Markdown tables. Merged cells, nested tables and tables that break across pages often lose rows or columns.
Can it read a scanned PDF?
Yes. Pages without a text layer are rendered to a canvas and read by a local OCR model. English and Simplified Chinese are available.
Are formulas converted to LaTeX?
No. Inline sub- and superscripts become $math$ spans, and detected display equations are kept as images in the ZIP download.
Is there a file size limit?
A single file can be up to 50 MB. Batch conversion is capped by the plan you are on; the beta is currently free.