DEV Community

Cover image for Which free online PDF tools upload your file? We tested 13 (Nov 2026)
Cuelark
Cuelark

Posted on Originally published at Medium

Which free online PDF tools upload your file? We tested 13 (Nov 2026)

By the Cuelark Team. Tested on 16 November 2026, between about 10:50 and 11:25 PM IST. Disclosure: we make Cuelark Tools, one of the 13 tools in this test. It was measured with the same script as everyone else.

Most free PDF sites promise your files are "secure" or "deleted after an hour". Fewer say plainly whether the file leaves your computer at all. You can check that yourself: open the browser's network panel and watch for a request about the size of your file.

We did that check with a script for 13 popular free tools, using one merge and one compress per tool and a harmless dummy PDF. This post gives the results, the method, what "upload" does and doesn't mean, and the limits of the test.

The short answer

Results: which tools sent the test PDF to a server

  • Sent the file to a server in every test we completed: iLovePDF, Smallpdf, PDF24 Tools, Sejda (web), Adobe Acrobat online, PDF2Go, Soda PDF, and Stirling PDF's hosted app (compress only, see below).
  • Mixed, depending on the task: Xodo and iHatePDF (ihatepdf.com) merged the two files inside the browser but uploaded the file for compression.
  • Kept the file in the browser for both tasks: ihatepdf.cv, BentoPDF and Cuelark Tools. Each produced a downloadable result, and none of the requests these pages made carried bytes from the test file.

In most of the tools that upload, the upload started as soon as the file was picked, before we pressed Merge or Compress. That was true for iLovePDF, Smallpdf, PDF24, Sejda, Adobe, PDF2Go and Soda PDF.

Results table

"Uploaded" means bytes from our test file appeared in a request after we picked the file (see the method below). The MB figure is the total sent to the host that received the file.

Tool Merge (2 files, 1.42 MB) Compress (1 file, 0.71 MB) Host that received the file What the tool says (official page)
iLovePDF Uploaded, 1.42 MB Uploaded, 0.71 MB api70.ilovepdf.com "all files processed within our platform are automatically and permanently deleted within two hours of being processed" (security page)
Smallpdf Uploaded, 1.42 MB¹ Uploaded, 0.71 MB Smallpdf's Cloudflare R2 storage bucket "Your PDFs are encrypted using TLS and automatically deleted from our servers after one hour." (merge page)
PDF24 Tools Uploaded, 1.42 MB Uploaded, 0.71 MB filetools0 / filetools26.pdf24.org "Files are automatically deleted from the PDF24 server after one hour" (merge page); "Servers are located in Germany"
Sejda (web) Uploaded, 1.42 MB² Uploaded, 0.71 MB² www.sejda.com "Files stay private. Automatically deleted after 2 hours." (merge page). Sejda's desktop app is described as one where "the files never leave your computer".
Adobe Acrobat online Uploaded, 1.42 MB Uploaded, 0.71 MB Adobe blob storage (adobe.io) "Your file will be securely handled by Adobe servers and deleted unless you sign in to save it." (merge page)
PDF2Go Uploaded, 2.84 MB³ Uploaded, 0.71 MB www3 / www11 / www14 / www27.pdf2go.com "The generated file can be downloaded a maximum of 10 times. Access is automatically blocked after 24 hours … At the latest, the files are physically deleted after 72 hours." (privacy policy, section 12)
Soda PDF Uploaded, 1.42 MB Uploaded, 0.71 MB api-gw.sodapdf.com "Your files remain private and will be deleted from our servers within 24 hours." (merge page)
Stirling PDF (hosted app, guest) Not tested⁴ Uploaded, 0.71 MB api.stirling.com Privacy policy covers hosted file content "per the retention settings you configure". We found no stated retention period for guest use (privacy policy).
Xodo Stayed in browser Uploaded, 0.71 MB Amazon S3 bucket "pdf-online-temp-storage" (us-west-2) Merge page: "Easily combine PDFs into one file in your browser". Compress page: "permanently deleted after 1h" (compress page)
iHatePDF (ihatepdf.com) Stayed in browser Uploaded, 0.71 MB⁵ ihatepdf.com "When you merge smaller PDF files, the whole job runs in your browser and nothing is uploaded." "Browser mode compresses on your device; files sent to our servers are deleted within an hour." (merge, compress)
ihatepdf.cv Stayed in browser Stayed in browser none "The merge runs inside your browser using pdf-lib, so your documents are never uploaded to a server." (merge page)
BentoPDF Stayed in browser Stayed in browser none "all file processing happens locally in your browser and we never collect or transmit your files to any server" (home page)
Cuelark Tools Stayed in browser Stayed in browser none "Nothing is sent to a server; you can confirm it in your browser's network panel while a file converts." (merge page)

Notes

  1. Our script didn't press Smallpdf's final merge button. Both files had already been uploaded when they were picked, and that upload is what we recorded.
  2. Sejda sends the file as a multipart upload labelled application/zip, so the raw PDF bytes don't appear as-is. The request was a single POST to /api/files/upload with filename test-a.pdf, about the size of the file. We count that as an upload.
  3. PDF2Go sent each file to one server and both files to another during the merge, which is why the total is about twice the input.
  4. Stirling's hosted app kept the two files in its in-browser file library when we added them, and no file bytes were sent at that point. Our script couldn't open the merge tool past the guest sign-up prompt, so we have no result for the merge step.
  5. iHatePDF's compress page offers a "Browser mode". We tested the page as it loaded, without changing any settings, and that run went to the server.

How we tested

Method: headless browser, logging proxy, tool servers

  • Test files. Two dummy PDFs we generated: 3 pages each, a line of text and a block of random pixels per page, 709,713 and 710,145 bytes. They contain no personal data. The random pixels make the files impossible to compress much, so an upload shows up clearly by size.
  • Browser. Chromium (Playwright) on a cloud Linux machine. Adobe's site refused the headless browser ("Access Denied"), so Adobe was tested in a regular, visible Chrome window instead.
  • Proxy. All traffic went through mitmproxy, which logged every request: host, path, body size, and whether the body contained any of 14 short (24-byte) samples taken from the test files.
  • Flow. Open the tool page, pick the file(s), wait 15 seconds, press Merge or Compress, wait 25 seconds, and press Download if a button appeared. A tool counts as "uploaded" if any request after the file was picked contained the file's bytes. Sejda (note 2) is the one exception, where we went by a same-size upload of the file.
  • Default settings only. No account and no sign-in. Cookie banners were accepted or declined only where they blocked the page.
  • Re-runs. We ran most tools at least twice. The tools that kept files in the browser were re-run with WebSocket logging turned on, and no WebSocket messages were sent from the page.

We kept the sanitized request logs (host, path without query strings, size, match count), screenshots and the test scripts. Ask and we'll share them.

"Uploaded" is not the same as "unsafe"

A server-side tool isn't automatically worse. Servers can do things a browser tab struggles with: OCR on scanned pages, high-quality conversion of Office files, heavy compression with tools like Ghostscript, very large files on weak phones, and saving work to an account. Most of the uploading tools mention an encrypted connection on the pages we read, and most give a deletion window from one hour (Smallpdf, PDF24, iHatePDF, Xodo compress) to two hours (iLovePDF, Sejda), 24 hours (Soda PDF), or 72 hours at the latest (PDF2Go). We didn't verify deletion, and nobody outside these companies can.

In-browser tools have trade-offs too. They depend on your device's memory and speed. Some features, like OCR and Office conversion, are harder to do well in the browser. And a page that processes files locally today can change tomorrow. If a document really matters (contracts, medical records, ID scans), the simplest rule is: either use a tool that doesn't upload and check the network panel yourself, or use desktop software.

How to check any PDF tool yourself in 60 seconds

  1. Open the tool, then open developer tools (F12, or Cmd+Option+I on a Mac) and go to the Network tab.
  2. Pick a file larger than about 1 MB.
  3. Sort by Size, or look for POST/PUT requests. An upload shows up as a request roughly the size of your file going to the tool's domain or a storage domain (*.amazonaws.com, *.r2.cloudflarestorage.com, and so on).
  4. For extra certainty, load the page, turn on airplane mode, then run the tool. If it still works, the processing happened on your device. Some in-browser tools load their PDF engine only when you pick a file, though, so they can fail offline even though they never upload anything.

Update, 17 November 2026: Cuelark Tools was one of those. Its merge and compress pages failed this airplane-mode test because the PDF engine only loaded once a file was picked. We confirmed this again on a copy of the old version. We changed the pages to fetch the engine quietly after the page finishes loading. We then re-ran the same script against the live site on 17 November at about 6:20 AM IST, going offline before picking files. Merge produced the merged PDF (1,417,040 bytes) and compress produced the 500 KB version (451,755 bytes). The engine files came from the browser's cache. The upload results in the table don't change.

Limits of this test

  • One day, one network, default settings. Results are from 16 November 2026, from a cloud server, with no account. Sites change often, and signed-in, paid, mobile app or desktop versions may behave differently.
  • Two tasks only. We tested merge and compress. The same site can upload for other tools (OCR, PDF to Word, e-sign) even if it merges locally, as Xodo and iHatePDF show.
  • Byte matching has blind spots. We search request bodies for raw samples of the file. A file that is compressed, encrypted or re-encoded before it's sent won't match (which is why Sejda needed a manual check), so we also looked at request sizes. Several sites sent analytics or error-monitoring requests during the tasks (Microsoft Clarity, Facebook, Reddit, Sentry). None of them contained matching file bytes. The largest was a 2.47 MB compressed payload to a Sentry host during the Adobe compress test. We don't know what was in it and we're not claiming it contained the file.
  • We don't check what happens after upload. Deletion windows are each company's own statement, quoted above.
  • No result for one test. Stirling's merge (note 4).
  • We're not neutral by birth. We build one of the tools in the table. That's why we published the method and kept the raw logs, so anyone can repeat the test and correct us.

If you run this test and get a different result, tell us and we'll update the table with the date.

Cuelark Team

Top comments (0)