Every tool in this section reads, changes and writes your document inside the browser tab you are already looking at. Nothing is uploaded. This page explains what that actually means, what a browser can do to a PDF, and where the honest limits are.
When you drop a file into an ordinary online PDF tool, the bytes travel to a server, wait in a queue behind everybody else, get processed, and come back. The service then promises to delete its copy — usually within some number of minutes. That promise may well be kept, but it is a promise about somebody else’s machine, and you have no way to check it.
These tools do not make that copy. The file is read by JavaScript running in your tab, changed in that tab’s memory, and written back out as a download. There is no request carrying your document anywhere, which is why the tools keep working when the connection drops and why there is nothing to sign up for: there is no server cost per file to recover.
The practical test is simple. Load the page, turn your network off, and use the tool. It still works. Nothing that needs a server can pass that test.
More than most people expect. A PDF is a structured document format, and reading and rewriting that structure is ordinary programming — merging, splitting, reordering, rotating, cropping, stamping text or an image onto a page, adding page numbers, and writing real AES-256 encryption all happen at the structural level. Nothing is re-rendered, so the output is exactly as good as the input.
Rendering is the other half. The browser can rasterise a page to a canvas at any resolution, which is what makes PDF-to-image, hard compression and burnt-in redaction possible. And it can run optical character recognition on that raster to recover text from a scan, using a recognition model that downloads once and then runs on your device.
A PDF that is already mostly text has very little to give up — the weight lives in images. Getting a large document under a fixed ceiling therefore means re-rendering its pages as pictures, and once a page is a picture, the words on it can no longer be selected, searched or copied. Every tool that does this says so before you download. The plain compressor also offers a lossless repack that keeps the text intact, and Smart mode picks whichever result is actually smaller.
Word to PDF and PDF to Word here read paragraphs, headings and lists and lay them out fresh. They do not reconstruct the original page-for-page design — tables, images and exact fonts do not survive the trip. For a letter or an essay the result opens like the source; for a document built around a table, Word’s own Save as PDF is the better answer, and it is worth saying so rather than letting you find out after the download.
There is no imposed file size limit, because there is no server to protect. What there is instead is memory: a very large document has to fit in the tab, and a phone runs out sooner than a desktop. Rendering forty pages is seconds of work and hundreds of megabytes of allocation, which is why the tools show a progress bar and let you cancel.
A scanned page is a photograph of words, not words. Anything that reads text — extracting it, counting it, highlighting it, redacting it, converting to Word, Excel or EPUB — has nothing to work with on a scan, and will say so rather than hand back an empty file. Running the document through OCR first adds the text layer those tools need.