How we build and test our converters
By Diego Callegari · Last updated October 2026
In short
Every tool runs in your browser, so files are never uploaded — you can verify that in two minutes with the steps below. Before every release, an automated test opens every tool in a real browser, converts sample files and checks the result is a valid file. Parsers for data exports have their own tests with sample data in the documented format.
Check for yourself that nothing is uploaded
Method 1: go offline
- Open a tool page, for example HEIC to JPG, and let it load.
- Turn off Wi-Fi (or switch on airplane mode).
- Drop a file and convert it. It still works — because the conversion happens on your device.
Some tools load an engine the first time you use them (for example text recognition or the video encoder); use the tool once while online, then try offline.
Method 2: watch the network
- On a computer, press F12 (or ⌥⌘I on a Mac) and open the Network tab.
- Drop a file into a tool and convert it.
- Look at the requests: you'll see the page's own scripts and anonymous statistics, but no request that sends your file. A 20 MB photo would show up as a 20 MB upload — there is none.
We ran this same check on six well-known online converters and on our own PDF to JPG page; the results are in Do online PDF converters upload your files?
What runs where
- PDF — pdf.js (the engine in Firefox) to read and draw pages, pdf-lib to create, merge, split and rotate them.
- Images — your browser's own decoders, plus libheif for HEIC, libjxl for JPEG XL and our own Ultra HDR gain-map handling.
- Text recognition — Tesseract, compiled to WebAssembly, with language models served from this site.
- Video and audio — the browser's built-in encoders (WebCodecs), often using your graphics card; LAME for MP3.
- Word, Excel and PowerPoint — docx and PptxGenJS, and our own .xlsx reader and writer and LaTeX-to-Word equation converter.
- Data exports — our own readers for Spotify, Netflix, YouTube, WhatsApp, Google Takeout, Apple Health, CapCut and more.
Each tool page names its engine under the tool, so you know exactly what processes your file.
How we test
Every tool, before every release
An automated test opens each of the site's tools in a real Chrome browser, drops sample files into it — photos, PDFs, videos, spreadsheets, chat exports — clicks the buttons a visitor would click and checks the download: that a JPG really is a JPG, a PDF opens, a ZIP contains files. A tool that fails is fixed before the release goes out.
The parts that are easy to get wrong
The tricky logic — reading dates in a dozen formats, counting messages per person, turning LaTeX into Word equations, keeping leading zeros when a CSV becomes Excel — has its own unit tests, including the odd cases we have run into: files with a byte-order mark, dates in the 1904 Excel system, time-zone offsets, exports in older and newer layouts.
Real files
Services rarely document their export formats. Each data reader is built from the format as it appears in real downloads and public descriptions, and is made tolerant of renamed columns and moved folders. When a format changes, the anonymous "Didn't work?" button on each tool tells us which kind of file wasn't recognised — only the file types and folder names the service itself uses, never file names or content — so we can fix it.
Limits we are honest about
- Your device does the work, so very large files (multi-GB videos, 3D scans) depend on its memory and speed.
- Some formats need a recent browser: video encoding works in current Chrome, Edge and Safari.
- Password-protected or DRM-protected files (some PDFs, e-books from stores) can't be converted.
- Recaps are made from what a service puts in your download; if a number isn't in the export, we don't guess it.
Found a problem?
Use the "Didn't work?" button on the tool, or contact me with what you tried. More about the site: About.