On-device scanning explained: how a browser scans without uploading

“Processed locally in your browser” sounds like marketing until you see the machinery. Here's what actually runs where when you scan with ToPDF — in plain language.

The one-time download (that’s the trick)

On your first visit, ToPDF downloads its engine: a computer-vision library and several small neural networks (document edge detection, page orientation, text reading), compiled to WebAssembly — a format browsers execute at near-native speed. You see it happen: the progress bar on the splash screen. From then on the engine lives in your browser's cache.

What happens to your photo, step by step

1. You pick a photo. It is opened in the tab's memory — the same way any image viewer opens a file.
2. The edge-detection model (running in the tab) finds the document's corners.
3. Perspective correction and enhancement are computed in the tab.
4. The orientation model turns the page upright; the OCR model reads a name for the file — all in the tab.
5. The PDF bytes are assembled in the tab and handed to your browser's download/share function.

At no step does the photo, the text or the PDF travel over the network. The only network activity is fetching the app itself — which is why everything keeps working with Wi-Fi off.

Trust, but verify

This isn't a promise you have to take on faith: airplane-mode the device and scan — it works. Watch the network tab in your browser's developer tools during a scan — no uploads appear. The architecture is verifiable from the outside, which is the standard privacy claims should meet. (Step-by-step verification guide.)

Frequently asked questions

If AI models are involved, don’t they need a server?

Small, specialised models don’t. Edge detection and orientation are a few dozen megabytes total and run in milliseconds-to-seconds on phone hardware via WebAssembly.

Does the engine download contain my data?

No — it’s the same static program files for everyone, cached like any website asset. Your content only ever exists in your tab’s memory and in files you save.

What does ToPDF’s server see, then?

Only ordinary web-hosting activity: that a browser fetched the site’s files. No analytics, no scan telemetry, no document content — there is no endpoint that accepts any.