Including the parts that will never exist
Kagaz has one architectural rule — your file never leaves your device — and that rule decides what can be built here and what cannot. This page lists all three categories, including the last one, because a roadmap that only contains good news is a marketing page.
Downsamples the images inside and leaves the text as text. Reports what it skipped and why.
All of these are possible in a browser. They are simply work, and this is a project built in evenings. Anything here is fair game if you would like to pick one up.
Add text, shapes and images to an existing page. The PDF side is straightforward; the editing interface is the real work.
Make a scan searchable. Fully possible in-browser, but the language data is 10–15 MB per language and it is slow on a phone, so it needs care to not feel broken.
Detect existing form fields, fill them, and flatten. The underlying library already supports it.
Side-by-side text and visual differences between two versions.
Text extraction with headings, lists and tables preserved. Quality will vary honestly by document, which is worth saying up front.
Partly works already — files with broken cross-reference tables often open here. A dedicated recovery tool would go further.
Genuinely possible but genuinely delicate: it means implementing PDF encryption by hand, and that is the kind of code that is subtly wrong for years. It will take as long as it takes. Kagaz will never attempt to guess a password.
Edge detection and perspective correction for photographed pages — the honest gap between Scan to PDF and a dedicated scanner app.
Find every occurrence of a name across a document and remove them all, rather than marking each one by hand.
A service worker so Kagaz installs like an app and survives having no network at all.
Hindi, Odia, Tamil, Bengali. This needs no programming at all — just someone willing to translate.
These are not oversights and they are not on a waiting list. Each one requires software that cannot run inside a browser tab, which means a server, which means uploading your document to a computer you do not control. That is the one thing Kagaz exists to avoid, so the answer is permanently no.
Other sites do offer them, and for a restaurant menu that trade may well be worth it. It is worth knowing you are making it.
A PDF describes where marks sit on a page. It does not know what a paragraph, a table or a heading is. Reconstructing an editable document means inferring all of that structure back — a heavy, model-driven job that no browser can do well. A bad version would produce a mess and quietly damage the trust the rest of this site runs on.
Converting the other way means laying out an office document exactly as its own application would — the same fonts, the same line breaking, the same pagination. In practice that means running a full office suite on a server. Anything short of it gets the layout wrong, and a contract that repaginates is worse than no conversion at all.
This needs the page to be fetched and rendered somewhere. A browser cannot fetch an arbitrary website from another origin, and the tools that do this run a headless browser on a server. For a page you can already open, your browser's own print-to-PDF does the job perfectly.
The standard requires embedded colour profiles, full font embedding and compliance validation against a specification. Producing a file that claims to be PDF/A without being able to verify it is worse than not offering it, because the whole point is that an archive can trust the label.
Both mean sending your document's contents to a model running on somebody else's hardware. There is no version of this that keeps the promise on the front page.
A legally qualified digital signature needs a certificate authority, a trusted timestamp and an audit trail — all of which are, by design, third parties. Sign PDF places a picture of your signature, which is what most everyday paperwork asks for, and it says so plainly rather than implying more.
Removing a password you know is a planned feature. Guessing one you do not is not a technical limitation — it is simply not something this project is willing to be.
“Two million documents processed” is good marketing, and it has to be counted somewhere — which means a request from your browser on every visit. Tiny, anonymous, and still this site watching you. So there is no counter, and I genuinely do not know how many people use Kagaz. The repository's own tests fail the build if a page ever gains a network call, which is what makes that claim checkable rather than a promise.
Including the self-hosted, cookie-free, privacy-respecting sort. All of them work by sending something about you somewhere, and the exception would be impossible to explain on a site whose whole argument is that nothing does.
Storing your files would require exactly the server this site does not have. There is nothing to log in to, and nothing to delete afterwards.
Roughly two thirds of what a full-featured PDF service offers can be done properly inside a browser. The third that cannot is almost entirely office-format conversion and AI — and every item in it fails for the same single reason.
That is the trade Kagaz makes. Fewer tools, and none of them ever see your document.
Open an issue and say what you were trying to do — the reason behind a request is usually more useful than the request. If you would rather build it, everything in the middle section is scoped and waiting.