PDF tools — private by design

Which tool for which job

The page operations share one property: pages are copied between documents, never re-rendered — text stays selectable, images keep their stored resolution, and the output is roughly the size of its parts. Merge joins several files in the order you list them. Split turns one file into one PDF per page, each with its own download button. Extract pages keeps only the pages you tap or type ("1, 3-17") together in one new file, in document order; Delete pages is its inverse.

Getting pictures out is two different jobs. PDF to JPG and PDF to PNG render whole pages at about 2,000 pixels on the long side, so the text in them becomes pixels. Extract images pulls out the pictures embedded inside the pages and re-saves each one at its stored pixel size, one file per distinct image — a logo repeated on every page comes out once — JPG when opaque, PNG when transparent. Going the other way, JPG to PDF, PNG to PDF, HEIC to PDF and TIFF to PDF put one image on each A4 page, portrait or landscape to match; JPG and PNG bytes are embedded as they are, HEIC and TIFF are decoded on your device and embedded losslessly.

Compress PDF does one precise thing: it downscales and re-encodes the JPEG images inside the document and touches nothing else. Light caps images at 2,400 pixels, Balanced at 2,000, Strong at 1,600; a re-encode is kept only when it shrinks that image by more than 5%, and if the finished file is not smaller than the original, the original comes back with a note saying so. Scans and photo-heavy reports routinely shrink 40–70%. PDF to text reads the text layer into one plain .txt. Rotate, Resize (A4, US Letter, Legal, A5, or a percentage) and Crop change page settings as vector operations — nothing is re-rendered.

Why these documents should never touch a server

Contracts, ID scans, medical records, bank statements: the files people run through PDF tools are the last things that belong on someone else's server. Here there is no upload and no account. The page you load is static code; once it is in your browser it reads your file, processes it, and writes the result on your device. There is no upload endpoint to send anything to.

You do not have to take that on faith. Every page ships a Content-Security-Policy that your browser enforces — a meta tag in the page source — with default-src 'none', form-action 'none', and a connect-src that allows the page's requests to reach only this site and the hosts of two analytics services (Cloudflare Web Analytics and Google Analytics 4, which count page views; see the privacy policy). The browser blocks any request from the page to any other destination. The simpler check: open your browser's developer tools, watch the Network panel while you merge or split, and see that no request carries your file. More on the security page.

Limits, stated plainly

• Encrypted PDFs. A password-protected document cannot be read without its password; the tools report it as unreadable rather than producing a broken file, and never ask for the password. Save an unlocked copy from your PDF reader first.
• Scans and OCR. PDF to text reads only the text layer. A scan has none, so the tool says so instead of handing you an empty file; OCR is not offered. If you can select the text in a viewer, extraction works.
• Layout. Text extraction flattens columns, tables, and fonts into a reading flow; line breaks stay, pages are separated by a blank line. To keep a page's look, render it with PDF to JPG.
• What compression reaches. Only JPEG images without transparency masks are re-encoded; PNG-style images, masked images, and vector artwork are left alone, so a file whose weight sits there shrinks little — and is returned unchanged, never inflated.
• Hiding is not redaction. Crop leaves the hidden content in the file, and Extract pages can carry fragments of a dropped page that a kept page links to. Use a real redaction tool for sensitive material.
• Memory. No page or size cap; the work runs in your device's memory. The tool warns before starting when a batch exceeds about 12% of the reported RAM (300 MB on a phone). Page pickers show thumbnails for the first 48 pages; type ranges beyond that.