Document converter

These tools turn Office documents into PDFs without sending them anywhere. The conversion is code running in your browser tab, which is possible because the modern Office formats are open and readable — and which means your payroll spreadsheet, your unreleased deck and your signed contract never leave your machine. Every one of these pages inspects the file you pick and names what it cannot reproduce before you spend a click on it.

A .docx is a ZIP archive of XML, and that changes what is possible

Rename any .docx, .xlsx or .pptx to .zip and open it. Inside is a documented folder tree: word/document.xml, xl/worksheets/sheet1.xml, ppt/slides/slide1.xml, plus styles, themes, numbering rules and a media folder holding the pictures. This has been an open standard, ECMA-376, since 2006. There is no proprietary blob at the centre of it.

The consequence is that reading an Office file needs a ZIP reader and an XML parser — both of which your browser already has — not a copy of Office. What is left to build is the part that knows what the tags mean and how to lay the result out on a page. That is real work, and it is where the fidelity differences between converters come from, but it is work that runs perfectly well on your own processor.

The older .doc, .xls and .ppt formats are genuinely different: binary compound files from the 1990s, far harder to read and not practically documented. None of these tools accept them. Any of Office, LibreOffice or the Google editors will re-save them in the modern format in two clicks, and that is the right first step.

Why other converters ask you to upload

Almost every online Word-to-PDF service runs LibreOffice on a server. You upload, a full office suite opens your file and prints it, and you get a download link. Fidelity is excellent because it is literally the same layout engine that would open the file on your desktop. The cost is that a machine has to run, a queue has to be managed, and your file has to sit on their disk for at least a few minutes.

That per-conversion cost is the reason those sites have the shape they do: a size cap, a limit of two files per hour, a watermark on the free tier, a paid plan that removes them. None of it is arbitrary. Server processing genuinely costs money, and rationing it is how the business works.

Doing the layout in the browser removes that cost entirely, and with it every reason to ration. There is no size cap here, no hourly limit, no watermark and no account, because a conversion costs us nothing. We also measured the speed difference: LibreOffice took 1.9 to 6.6 seconds per test document on a local machine; the same conversions took 13 to 353 milliseconds in a browser, with no upload and no queue in front of them.

Where layout fidelity actually breaks, and how to predict it

Fidelity does not degrade evenly. A single column of text with headings, lists, tables and PNG or JPEG images comes out very close to the original — that covers résumés, letters, reports, invoices and most technical documents. What breaks is structure that is not a single flow: multi-column sections get flattened to one column, text boxes keep their words but lose their position, headers and footers are not written, and paragraph borders are not drawn.

Three things are not converted at all, deliberately: charts, SmartArt and equations. Each is effectively a separate rendering engine, and a chart that is nearly right is worse than a blank space because it still looks like data. Spreadsheets have their own version of this — images are not drawn in the current version, and formulas are never recalculated, which matters because a file that stores a formula with no saved result would otherwise print a plausible zero.

You do not have to predict any of this from experience. Each tool reads your specific file first — counting the columns, text boxes, charts, images and every character it will have to draw — and gives a verdict with named reasons in a few milliseconds, before converting anything. A file that will come out wrong is labelled as such, and the button still works, because deciding for you that you may not try is worse than letting you try once.

Chinese, Japanese and Korean: what works and what we refuse

A PDF can only draw a character if the drawing font is embedded in it or is one of the fourteen standard fonts every reader has. Those fourteen are Latin-only, so Traditional Chinese needs a real embedded font. These tools ship a 6,403-character subset of Noto Sans TC — common characters, punctuation, bopomofo, full-width forms, box drawing, circled numbers — fetched only after you choose a file and only if that file needs it. Around one real Chinese document in five uses a rarer character, usually one or two of them and often in a name; those files get a larger extended set instead.

Japanese and Korean are refused, with the reason stated. This is not a size decision. Noto Sans TC contains none of the 11,172 Korean syllable characters, and a real Korean sentence renders at 9.5% even with the complete source font. Japanese would be worse in a specific way: adding kana would still leave common Japanese-only kanji at about 70%, so pages would come out half correct — harder to explain than a clear refusal — and our detector would stop warning about exactly those files. We tested that outcome before deciding.

The bundled font is under the SIL Open Font License 1.1 and the licence text ships with it. Section 5 of that licence states that the requirement to remain under it does not apply to documents created with the font, so the PDFs you produce here carry no licence obligation of their own. One thing we do not claim: these are not PDF/A files. Latin text uses the standard PDF fonts, which are not embedded by definition, and that alone puts the output outside the archival standard.