This is a re-save, not Acrobat Distiller
Upload farms advertise “90% smaller” because a server runs Ghostscript and recompresses every JPEG inside the file. I do not run that pipeline. This page loads pdf-lib from jsDelivr so the library can exist at all, then re-saves your document in this tab with object streams turned on. Watch the compute log: you should see a library fetch, not a POST of your deck. If the file host is blocked, the button cannot even start. That is still not your PDF leaving the device.
The picker refuses anything over 10 MB. That guard is in the script, not a mysterious cloud quota. A mid-range phone can still freeze under 10 MB if RAM is already full of other tabs. Password-protected PDFs usually fail to load. Digitally signed statements become a new file if a re-save succeeds; the signature will not remain valid, and I will not pretend otherwise.
Worked example: 40 MB scan pack
You export a board pack from a scanner. The picker says no. That is the 10 MB cap doing its job. Split the range with merge & split, then try each piece. If what you actually need is JPEG downsampling of 200 raster pages, a desktop compressor or a printer driver’s “smallest file” export will beat this tab. I would rather bounce you than ship a fake 90% claim.
Worked example: 8 MB text contract
A Word-to-PDF letterhead with vector fonts and one logo. Compress. The result might be 7.6 MB. Object streams tidy some waste. They do not delete type. If the original was a bloated print-to-PDF with unused fonts, you may see a real drop. If it was already lean, celebrate the 400 KB and send it. Staring at the quality slider until the number magically halves is how Saturday afternoons die.
Worked example: the quality slider barely moves the bytes
The control is there. The save path still does not walk every embedded photo and re-encode it the way Distiller does. If you drag quality and the size stays put, believe the size. For a file that is literally photographs-as-pages, rasterize with PDF to image, shrink the pictures in the image compressor, then restack with image to PDF. That round trip is lossy and honest. Preview before you email a client a mushy stamp.
What actually breaks
Encrypted PDFs. Files named .pdf that are HTML printouts. XFA forms. JavaScript inside the PDF. Linearized / fast-web-view hints. Bookmarks may survive a re-save better than a merge, but I still treat the output as a new document. Keep the original. If legal needs the signed bytes, send those, not this download.
One file at a time. Batch “drop a folder” belongs in a desktop app. Using an upload farm for passport scans is how leaks happen; using this tab for a 500 MB phone dump is how tabs die. Pick the failure you can live with.
If the download is larger than the original, that happens when object streams add wrapping to a file that was already linearized and tight. Keep the original. Compare byte counts before you replace the attachment in an email thread. A “compressed” 8.1 MB file that started at 7.9 MB is not a win; it is a new document with a new hash, which matters if someone is checksumming a packet. I would rather you send the original lean export than a worse re-save with a hopeful filename.
Questions
Is the PDF uploaded to Privy Tools?
No. Bytes stay in this tab. pdf-lib loads from jsDelivr so the tool can run; that hop is the library, not your file.
Why did the file barely shrink?
Text and vector PDFs have little image waste. This path re-saves with object streams. It is not Ghostscript JPEG recompression.
What is the size limit?
The picker caps at 10 MB. Device RAM can fail earlier, especially on phones.
Password-protected PDFs?
Usually fail. Decrypt in a tool you already trust if that is lawful, then compress the unlocked copy.
Will a digital signature survive?
No. A re-save is a new document. Keep the signed original for anything that must stay signed.
Does this OCR the file?
No. Compression does not make a scan searchable. Use a dedicated OCR tool if you need text.