Turn messy PDFs into clean Markdown your AI can actually read.

1. Step 1

   Drop your PDF

   Drag the file onto the box above, or click to choose one. It stays on your computer. Nothing is uploaded, so there is no waiting for a server.

2. Step 2

   It rebuilds the document

   Columns are put back into reading order, tables are kept as tables, headings stay headings, and repeated page furniture is stripped out.

3. Step 3

   Copy or download the .md

   Take the clean Markdown straight into ChatGPT, Claude, a RAG pipeline, or your notes. You also see roughly how many tokens you saved.

The options for choosing a customised
interior design are 1. Brown/Light
practically limitless. To make your
choice a little easier, our 2. Brown/Dark
designers have put together a selection
of interiors with 3. Black/Dark
perfectly matched colours.

The options for choosing a customised
interior design are practically
limitless. To make your choice a little
easier, our designers have put together
a selection of interiors with perfectly
matched colours.

1. Brown/Light
2. Brown/Dark
3. Black/Dark

Left is what a scrambled read produces: option labels dropped into the middle of sentences. Right is the same page after reading order is reconstructed. Feed the left one to a model and it will answer from nonsense.

This is a single real document, not an averaged benchmark suite. Your own reduction depends on how much of your PDF is layout overhead versus text. The tool shows you the before and after on every file, so you never take our word for it.

Is my document uploaded to a server?

No. Conversion runs entirely in your browser using pdf.js. The file never leaves your device, and nothing is sent to Siteiz or any third party. That makes it safe for confidential contracts, financial statements, and other regulated documents.

Does it support scanned PDFs or images (OCR)?

Yes. Pages with no text layer are detected automatically and read with built-in OCR that runs in your browser using Tesseract, an open-source engine. The engine downloads on first use, and your document still never leaves your device. OCR is part of PDF Pro, a 99 kr pass that covers unlimited scanned pages on any file for 24 hours.

What does it cost?

Converting text-based PDFs is free and unlimited, with no account and no sign-up. Scanned PDFs need OCR, which is the PDF Pro pass: 99 kr for 24 hours of unlimited OCR on any file. It is a single charge that simply stops after 24 hours. Nothing renews and you are not charged again.

How long does the 99 kr pass last?

24 hours from the moment your payment clears, not from first use. During that window you can convert unlimited scanned pages on any number of files. When the 24 hours are up, OCR stops and free text conversion continues as normal. If you need OCR again later, you buy another pass. Teams that need OCR continuously are better served by the annual business licence.

I bought a pass. What happens if I clear my browser data?

Your pass is stored in your browser rather than in an account, so clearing site data removes it from that browser. When you buy, the page shows a pass link: save it, and opening that link restores your pass on any browser or device for as long as the 24 hours are still running. Lost it? Email hello@siteiz.com from the address you paid with.

Do you offer a business or team licence?

Yes. The business licence is 20 000 kr per year, billed annually by invoice. It covers unlimited OCR and native processing for commercial use across your organisation, gives your team a shared access link with no user registration, and includes a direct line to support for custom formatting or roadmap requests. Because conversion happens entirely in the browser, your documents stay on your own machines, which is usually the reason regulated teams get in touch. Email hello@siteiz.com to request a corporate invoice.

How accurate is OCR on scanned documents?

OCR is a best-effort reading of the page image, so proofread anything critical. Clean, high-resolution scans of printed text read well. Low-resolution faxes, handwriting, heavy background imagery, and unusual typefaces are harder, and pages that are mostly photographs can produce stray characters.

Is there a file size limit?

Because processing is local, the practical limit is your device memory, not a server quota. Text PDFs of several hundred pages convert comfortably on a typical laptop. Very large or image-heavy files depend on available RAM.

How do I parse multi-column PDFs for LangChain or LlamaIndex?

Convert the PDF to Markdown here first, then load the .md with your framework's Markdown loader instead of a PDF loader. Most PDF loaders read straight across the page, so a two-column paper arrives with its columns interleaved mid-sentence and every chunk is corrupted. Reconstructing reading order before chunking is what makes retrieval return coherent passages.

Does this work for RAG pipelines and embeddings?

That is what it is built for. Markdown headings give recursive character or Markdown-header splitters real boundaries to chunk on, so a chunk ends at a section rather than mid-sentence. Tables stay as tables, so row and column relationships survive into the embedding. Repeated page headers and footers are stripped, so the same boilerplate is not embedded hundreds of times.

Can I convert PDF tables to Markdown without uploading data?

Yes. Table detection and the whole conversion run inside your browser, so no data is transmitted. Tables are found by clustering glyph coordinates into rows and columns and are only emitted as a Markdown grid when the structure is confident; ambiguous layouts fall back to clean text rather than a broken table.

Why Markdown instead of plain text for LLMs and RAG?

Markdown preserves the structure that plain text discards: headings, lists, and tables. That structure improves chunking, retrieval accuracy, and citation quality in RAG pipelines, while staying compact and token-efficient.

How is this different from other PDF to Markdown converters?

Most extractors read straight across the page, scrambling columns and flattening tables. Siteiz reconstructs the document with reading-order segmentation and confidence-gated table detection, and it does all of it locally, so structured output and data privacy come together instead of as a trade-off.