PDF Desk

Merge, split, rotate and reorder PDF pages in the browser. Nothing is uploaded, nothing is installed, and the page works offline once it has loaded.

Live: https://freehtml.dev/pdf · Source: https://github.com/TekiDeki/PDF-Desk

What it does

• Merge any number of PDFs, JPGs and PNGs into one file, in the order you choose.
• Reorder files, and reorder pages inside them.
• Rotate a page a quarter at a time. Rotation is added on top of whatever the page already had, so a scan that was sideways in the original comes out straight.
• Delete pages you do not want in the result.
• Extract a page range, in your own order, with repeats if you want them, for example 1-3, 5, 8-10 or 4, 1, 1.
• Turn images into pages: one image per page, on A4, on Letter, or at the picture's own size.
• Undo, twenty steps back, for anything done to the page list.
• Dark theme, remembered between visits, or the system one on the first visit. The page tiles stay white on purpose: they stand for the document itself.

Everything happens in the tab. There is no server, no account and no analytics.

What it does not do

• No password removal. Encrypted PDFs are refused, not cracked.
• No text editing. Pages are moved and turned, never rewritten.
• No OCR. A scan stays a picture.

These are limits, not bugs. They are also on the page itself, so nobody is surprised after the work is done.

Using it

1. Drop PDFs or images on the box, or press Choose files.
2. Fix the order: arrows on each file, click a page to turn it, cross to delete it.
3. Press Merge and download. Or write a range and press Download these pages.

The merge and the extract both start from the page list you see on screen, so the extract can be used to save a corrected copy without downloading the whole document first.

Development

python3 build.py                  # writes index.html, single file, deterministic
node tests/logic_test.cjs                      # 68 checks: engine and real PDF work
python3 tests/ui_test.py                       # 75 checks on Chromium
BROWSER=firefox python3 tests/ui_test.py       # the same on Firefox
BROWSER=webkit  python3 tests/ui_test.py       # the same on WebKit, the Safari engine

Every pass covers a desktop, a touch tablet at 820x1180 and a touch phone at
390x844, and every download is read back through pdf-lib and counted. The same
flow has been run against the live site in all three engines.

tests/logic_test.cjs needs the library installed next to the repo:

npm install pdf-lib@1.17.1

The UI test starts its own static server, so nothing else has to be running.

Files

Path	What it is
src/template.html	the page skeleton, with the three script slots
src/styles.css	all the styling, no framework
src/engine.js	pure functions plus the two PDF operations, no DOM
src/app.js	the interface: files, pages, language, downloads
src/vendor/pdf-lib.min.js	pdf-lib 1.17.1, MIT, inlined at build time
src/i18n/locales.json	EN, RO and SR Latin; SR Cyrillic is generated
build.py	inlines everything into index.html and checks the rules
index.html	the built page, committed, served by GitHub Pages
sw.js	offline shell
screenshots/	desktop, phone, empty and the dark theme, from this build
tests/fixtures/	PDFs and images used by the tests, including one encrypted and one corrupted on purpose

Languages

Four: English, Romanian, Serbian Latin and Serbian Cyrillic. Romanian and Serbian Latin are written by hand; Serbian Cyrillic is produced by the build from the Latin one, so the two can never disagree. The build refuses a missing key, a placeholder that does not match between languages, an em dash, an en dash or a byte order mark.

Language detection uses navigator.language, the choice is remembered in localStorage, and the page name is kept there too.

Limits worth knowing

• A very large PDF can run out of memory on a phone. The page says so instead of failing quietly. Measured on this build: three 20 page scans of 4.7 MB each (14.1 MB of output, 60 pages) merge in about 0.4 s in a phone sized window.
• The page shows the first 400 pages as tiles. Every page is kept in the file, the cap is only about drawing them.
• Images are placed whole, without cropping or margins control.

Licence

Apache-2.0. The bundled pdf-lib is MIT, by Andrew Dillon; its notice is in src/vendor/.

Built by TekiDeki. Free, no account, no tracking. If it saved you some time: https://paypal.me/TekiDeki