Before you edit a PDF, find out where the file is going
The documents people most need to edit are the ones they least want to upload. Most PDF tool comparisons never mention where the processing happens.
Published · Keenvale
Think about the last five PDFs you needed to change. A signed contract. A payslip. A scan of an ID for a form. A medical result. A tax document.
Now think about what you did: searched for “merge PDF free”, clicked the first result, and dragged the file into a web page.
That is the entire subject of this guide. Not which tool has the most features — whether the file left your computer, and whether you would have chosen that if anyone had asked you.
PDF is unusual this way. Nobody uploads their photo library to a stranger’s server to rotate an image. But the PDF is the format we reserve for exactly the documents that identify us, pay us and bind us, and it is also the format with the largest ecosystem of free web tools asking us to hand the file over.
Three places the work can happen
Every PDF tool does its processing in one of three places, and the feature list will not tell you which.
In your browser tab. Some web tools do the work in JavaScript on your machine; the file never crosses the network. This is genuinely private and also genuinely limited — heavy operations like OCR strain it.
On your machine, in an application. A desktop program, or a service you run yourself. The file stays put by construction.
On someone else’s server. You upload, they process, you download. This is what most free web tools do, because it is the only way to offer heavy features for free in a browser. It is not sinister — it is a design constraint. But it means a copy of your document existed on hardware you do not control, governed by a retention policy you did not read.
How to tell which one you are using
Not from the marketing page. Three checks that work:
- Try it offline. Disconnect from the network and run the operation. If it completes, the work was local. This takes ten seconds and is the most reliable test there is.
- Read the retention sentence, not the privacy page. You are looking for a specific claim about how long uploaded files are kept and when they are deleted. A privacy policy that talks at length about cookies and never mentions document retention is telling you something.
- Ask what happens to a 300-page scan. If a tool offers free OCR on large documents in a browser, the processing is almost certainly not happening in your browser.
What we can verify about the tools people actually use
Stirling PDF — the self-hosted option
Stirling PDF is an open-source PDF platform, open-core by its own description, with around 89,000 stars on GitHub at the time of writing. It runs as a desktop application, in the browser, or deployed on your own server.
Its central claim is the one this guide is about: you can “edit, sign, redact, convert, and automate PDFs without sending documents to external services.” On file handling, the project states that files “exist either exclusively on the client side or reside in server memory only during task execution”, with temporary files deleted after use.
Installation is a single Docker command if you already run Docker, and there is a desktop path if you do not. It advertises 50+ tools including merge, split, redact, OCR and compression.
The honest caveat: self-hosting is a real commitment, not a free lunch. You become responsible for updating it. For someone who edits a PDF twice a year, that trade is probably bad. For someone handling client documents weekly, it is probably excellent.
Adobe Acrobat — and a price we are not going to state
Acrobat is the reference implementation, and the desktop application does its work on your machine. Adobe also runs free online tools, which are server-side by construction — the same product name covers both models, which is worth knowing before you reach for whichever is closer to hand.
On price, we are going to do something unusual for a comparison article and decline to give you a number.
We tried. Adobe’s pricing page renders its figures dynamically and varies them by region, and we could not extract a stable, quotable price for Acrobat Standard or Pro for individuals. Every comparison article you will read confidently states one. Some of them are quoting a US list price at a reader in Brazil, and some are quoting a promotional rate that expired.
So: open Adobe’s pricing page yourself, in your own country, and read the number there. That is the only figure that applies to you, and it takes fifteen seconds. We would rather send you to the source than publish a number we could not verify.
What actually decides this
| If you are… | The right answer is usually |
|---|---|
| Editing an occasional non-sensitive document | A browser tool is fine. Just be deliberate that it is non-sensitive |
| Handling contracts, IDs, payroll or medical files | Local processing, without exception. Desktop app or self-hosted |
| Doing this weekly for other people’s documents | Self-hosted, and worth the setup cost |
| Needing OCR on large scans, occasionally | This is the genuinely hard case — heavy processing pushes you toward a server or toward paying for a desktop tool |
The redaction warning that belongs in every PDF guide
One thing worth more than any tool choice: drawing a black rectangle over text is not redaction. In many tools the text remains in the file underneath the shape, selectable and copyable. Real redaction removes the underlying content, and tools that do it correctly call it “redact” rather than “annotate” or “mark up”.
If you have ever blacked something out on a PDF and sent it on, that is worth checking today. Open the file and try to select the text under the box.
What we have not tested yet
Everything above about Stirling PDF comes from the project’s own repository and documentation, quoted rather than paraphrased. We have not yet run it, timed an OCR job, or verified the offline test on each tool ourselves — which is, by this guide’s own argument, the check that counts.
We also have not verified the redaction behaviour tool by tool. We know the failure mode is real and common; we have not yet built the table of which specific tools get it right. That table is worth more than anything else on this page, and we would rather publish it when we have actually tested it than guess.
We will update this page when we have.