100% private — processed in your browser with WebAssembly
PDF, Word, Excel, PowerPoint and images become clean Markdown for ChatGPT, Claude, Obsidian and Notion. Your files never leave your device.
Structure is context. A heading tells a model that what follows is a section on that subject, and a table keeps every number attached to its column header. Flatten a document into plain text and both are lost. Markdown keeps them while staying compact, so more of your document fits in a limited context window.
It is also plain text you can read and edit yourself before handing it over — which means you can delete the boilerplate or fix a mangled table first.
Drag in a PDF, Word, Excel or PowerPoint file — or a scan or screenshot.
Your browser does the work with WebAssembly. Nothing is uploaded, so there is no queue and no file sitting on someone else's server.
Take the Markdown straight into a ChatGPT or Claude Project, Obsidian, Notion, or a docs repo.
The same page, before and after. Headings stay headings and the table stays a table — which is the whole reason a model reads Markdown better than a PDF.
# Quarterly Results
## Revenue by region
| Region | Q3 | Q4 |
| -------- | ---- | ---- |
| Europe | 1.2M | 1.4M |
| Americas | 2.0M | 2.3M |Every conversion has a Chunks tab. It splits the document on its own headings, keeps the heading path on every piece so a chunk still means something when it comes back on its own, and never cuts through a table or a code block — a long table is split between rows with its header repeated, because the usual reason a converted spreadsheet retrieves badly is that only the first piece knows what the columns were.
{
"id": "report-0002",
"headings": ["Quarterly Report", "Revenue", "North"],
"tokens": 412,
"text": "Both northern offices reached capacity…"
}Download it as JSON and feed it to your loader. Getting a folder of documents into a vector store otherwise means a command-line tool or handing the files to an API — which is the upload this converter exists to avoid. This runs where the rest of it runs: in the tab, on your machine.
No. The conversion runs inside your browser and the file never leaves your device. There is no upload step, so there is nothing for us to store or delete. The privacy page explains exactly what is and is not collected.
In your browser, on your device. The parsing libraries and the OCR engine are downloaded to your machine and run there as WebAssembly, which is why there is no queue and no upload — and why it keeps working once the page has loaded.
Yes. If a PDF has no text in it, MDifying offers to read the text off the page images, and it does that locally too — the language models are served from this site rather than a third party. It recognises Spanish and English. Reading images is slower and less accurate than reading real text, so check the result before relying on it.
It is standard Markdown, with real tables and headings, so it drops into Obsidian,
Notion, MkDocs, Docusaurus or any editor that reads .md.
Yes — it produces standard Markdown, which you can paste into a conversation
or attach to a ChatGPT or Claude Project as a .md file. MDifying is not affiliated with either, and it does not
send anything to them; what you do with the converted text is up to you.
Yes — 50 MB per file on a computer, 10 MB on a phone. These are limits of what a browser tab can hold in memory, not account limits: the number of conversions is unlimited and there is no sign-up.
Yes, with no account and no conversion limit. Conversions cost nothing to run because they happen on your machine, not ours.
Something missing, or converted badly? — it decides what gets built next.
MDifying is free and has no ads, because conversions run on your machine and cost nothing to serve. If it saved you time, you can buy me a coffee.
Opens an embed from ko-fi.com. Nothing is loaded from them until you click.