What Browser-Only Processing Protects
Every tool on this site that touches a file does the work in the page you have open. That is a real property with a real benefit, and it is routinely oversold across the whole category of browser-based file tools. So here is the narrow claim, what supports it, and the list of things it very deliberately does not cover.
The narrow claim
When you pick a PDF, the browser's File API hands the tab a readable handle to the bytes on your disk. Those bytes are read into the tab's memory. A library running in that tab — pdf-lib for structural work, pdf.js for rendering pages, JSZip for packaging multiple outputs — rewrites them. The result is assembled as a Blob, given a temporary object URL, and handed to the browser's download machinery.
At no point in that sequence is there an HTTP request carrying your document. There is no server that receives it, no queue it waits in, no temporary directory it lands in, and no retention window to trust anybody about. The file never becomes somebody else's problem because it never becomes somebody else's file.
That single property is what most people are actually worried about when they hesitate over a tax return, a bank statement or a scan of a passport. It is worth having. It is also the entire claim.
Check it rather than believing it
You do not have to take this on faith, and you should not take it on faith from any site.
Open your browser's developer tools before you start, switch to the Network panel, then run the job. You are looking for an outbound request whose payload is roughly the size of your file. On a page that genuinely processes locally there is nothing — the code and fonts load, and then the network goes quiet while the work happens.
Two caveats on that method, because a check presented as proof is worse than no check. Something determined to exfiltrate a file could send it slowly in small pieces, and picking that out of ordinary traffic takes more than a glance at the panel. And the check tells you what happened during that one run on that one day. It is a good sanity test and it is not an audit.
The four things it does not mean
It does not mean offline. The page arrives over the network like any page, and the heavy libraries are not in the initial bundle — each one is fetched on first use, so the code for the watermark tool downloads at the moment you first watermark something. Disconnect after loading the site and any tool you have not already used will fail to fetch what it needs. There is no service worker here and nothing to install, so nothing has been cached for later on purpose.
It does not mean nothing else is running on the page. This site loads Google Analytics and Google AdSense, and it loads them only after you have accepted in the consent banner. Decline and neither script is injected at all. When they do load, they run in the same page as the tool. They cannot reach your file — the File object and the Blob are held in the page's own JavaScript scope and there is no mechanism by which another script obtains them without the code deliberately handing them over — but they do observe that a browser visited this address. That is a different kind of privacy from document privacy, and conflating the two would be dishonest.
It does not mean encrypted. Bytes sitting in a tab's memory are plain bytes. The output that arrives in your Downloads folder is a plain file that anything on the machine can open. The one place real cryptography enters is Protect PDF, which encrypts the document itself with a key derived from your password. Local processing and encryption are unrelated properties and having one says nothing about the other.
It does not defend you against your own machine. A browser extension granted access to page content is inside the same boundary as the tool. So is anyone else who uses a shared login on that computer, or an employer's management software on a work laptop. If your Downloads folder syncs to a cloud drive, the output you just kept off one company's servers is now on another's. Shared or managed devices are the weak point, not the network.
What it costs
Local processing is not free, and the costs land on you rather than on us.
Everything happens in tab memory, which is why there are size ceilings — 100 MB for a single file — and why a phone will kill the tab on a very large render long before a server would have struggled. Rendering is slow because it uses your processor. There is no optical character recognition, because a serviceable OCR model is not something to download into a browser tab. A server-side service with a rendering farm behind it can do things these pages cannot, and for some jobs that is the correct trade.
The honest summary is a single sentence with a hard boundary in it: your document is not uploaded and no copy of it exists on our infrastructure, and every other privacy question you have is about your browser, your extensions, your machine and your download folder.
If you want to see the shape of it yourself, Merge PDF is the simplest case — pick two files, watch the Network panel stay quiet, and check the result.