Verify it yourself: nothing you drop here is uploaded
You shouldn't have to take our word for it. Every claim about local processing on this site can be checked in under a minute, with tools already built into your browser.
Below: how to verify the engine yourself, the complete list of network requests this site makes (including the ones that leave this domain), the open-source engines behind each feature, and the limitations we admit to.
1. Three ways to verify it yourself
Offline test: load a converter page once, then switch your device to airplane mode — or tick “Offline” in DevTools → Network — and convert a .docx anyway. It still works, because the engine already runs on your machine. A tool that needed our servers would fail instantly.
Network panel: open DevTools → Network, clear the list, then convert a file. You will see the same-origin engine downloads (/pandoc.wasm, /pdf-inspector.wasm, /pdf.worker.min.mjs) and nothing else. Filter by Fetch/XHR — an upload would appear there as a POST to a foreign domain. There is none.
No backend exists to send anything to: this site is a static bundle of HTML, JavaScript and WebAssembly files. Search the page source for an upload endpoint or an API key — there is neither. If you ever find a request that carries your document, that settles the argument against us.
2. Every request this site makes
We would rather list the exceptions than have you discover them. The engine binaries are served from this domain and then cached by your browser: /pandoc.wasm (about 56 MB, the real Pandoc compiler, downloaded once), /pdf-inspector.wasm (about 6 MB, the PDF engine) and /pdf.worker.min.mjs (about 1 MB, the pdf.js worker used for OCR).
Three kinds of assets come from the jsDelivr CDN instead: the tesseract.js worker, core and language model, fetched only when you run OCR and only for the language you select; KaTeX (~340 KB) to render math in the HTML preview and in Markdown → PDF; and Mermaid (~1 MB) only if a document contains a Mermaid diagram block.
None of these requests carry your document or its contents. They are plain GET requests for library code and OCR language data. You can confirm it in DevTools: select the request and check that Payload / Request body is empty.
3. What actually runs in your browser
Word, ODT, EPUB, HTML, reStructuredText, LaTeX, CSV and more are converted by Pandoc compiled to WebAssembly — the same engine as the command-line tool, running in a background thread. PDF input goes through pdf-inspector (Rust → WebAssembly) and pdf.js; scanned or photographed text is recognised by tesseract.js. Markdown → PDF uses your browser's own print-to-PDF dialog.
Every engine is open source and is listed with its licence on the Terms page: Pandoc and the vendored pandoc-wasm core are GPL, pdf-inspector is MIT, pdf.js and tesseract.js are Apache-2.0. Because these engines are shipped to your browser, the corresponding source is the project linked there.
There is no proprietary conversion service in the middle, no metered API key and no account. That is why batch conversion and the Pro options (citations, Lua filters, custom templates) keep working with the network cable unplugged.
4. Honest limitations
The first visit downloads the engine: about 56 MB of WebAssembly, cached afterwards. On a slow connection the first conversion can take a while — a server-based converter would hide that from you; we cannot.
Offline only works after that first visit. A brand-new device with no network at all cannot convert anything, because the engine has never been cached.
OCR fetches its language model from the jsDelivr CDN the first time you use a language (about 2–15 MB each) and is far slower than native text extraction. Scanned pages, handwriting and low-contrast photos are the hardest cases.
Conversion between document formats is lossy by nature: complex layouts, columns, text boxes, tracked changes, comments and some equations do not survive every path. Always review the output before relying on it.
Word → PDF is not offered. PDF output through LaTeX is a Pandoc command-line feature, and Markdown → PDF is limited by what the browser's print dialog can produce — no font embedding options, no PDF metadata editing.
There is no REST API, no CLI and no automation hook. If you need server-side conversion at scale, run Pandoc locally: it is the same engine this site uses.
5. Still not convinced? Reproduce the checks
From a terminal: curl -s https://wordtomarkdown.com/en/ | grep -oE 'https://[a-zA-Z0-9.-]+' | sort -u lists every host referenced by that page. You will see this domain and https://schema.org — the latter is a JSON-LD identifier read by search engines, not a request — and no upload endpoint.
In DevTools, open the Network tab, convert a file, then click any request and inspect its Payload tab. An empty payload on every request means nothing was sent: the only traffic is downloads of code and models.
If you find a request that contradicts this page, tell us through the Contact page. We will correct the page rather than argue.