Every OCR tool promises privacy. Almost all of them mean the same thing by it: they'll send your image to a server, read it, and then not keep it around for too long. That's a policy, and policies are only as good as the company enforcing them, the breach that hasn't happened yet, and the terms nobody read.
There's a stronger version of the promise, and it's the one worth holding out for: the image never left your device in the first place. That's the guarantee Textquill is built around, and the difference between the two is not marketing. It's architecture.
Two very different meanings of "private"
Put the two models side by side:
| Claim | What actually happens | What can go wrong |
|---|---|---|
| "We don't store your images" | Image is uploaded, processed on a server, deleted after | Logs, caches, backups, breaches, policy changes, subpoenas |
| "Your image never leaves your device" | Model runs locally; nothing is uploaded | Nothing to leak, because nothing was sent |
The first row is a promise about behavior. The second is a fact about data flow. When the data never moves, the entire category of "what if the server does something with it" stops existing. You're not trusting a retention policy. There's nothing to retain.
This is the same distinction I drew in why on-device OCR is the smarter default: local inference isn't a privacy feature bolted on, it's what removes the risk at the root.
What the screenshot-to-text flow looks like
The reason people reach for OCR dozens of times a day is that text is constantly trapped in the wrong format: a screenshot a colleague pasted, an image-only PDF, a chart label, an error message someone photographed instead of copying. The flow you want is short:
- Capture the region or point at the image.
- Read it locally, in the browser.
- Get selectable, editable text you can paste anywhere.
With Textquill, that middle step happens on your machine. No upload, no spinner waiting on a queue, no "processing on our secure servers." The pixels go into a model that's already sitting in your browser, and text comes out. Because there's no network hop, "in seconds" is literal rather than aspirational.
Why this matters more at work than at home
The privacy argument gets waved off as paranoia until you think about what people actually OCR during a workday:
- Invoices and receipts with account numbers and totals.
- Support tickets and DMs with customer names and emails.
- Internal dashboards and admin panels that were never meant to leave the network.
- Contracts, IDs, medical forms captured as image-only PDFs.
Every one of those is data your employer, your customers, or a regulation has an opinion about. A cloud OCR tool turns each quick copy-paste into a quiet data-transfer decision, usually to a vendor nobody vetted. On-device OCR takes that decision off the table entirely. There's no vendor to vet because there's no transfer.
Handling anything you wouldn't want sitting in a stranger's server logs? That's exactly where Textquill earns its place: the screenshot becomes editable text, and the screenshot itself goes nowhere.
How to tell if OCR is really local
You don't have to take anyone's word for it, including mine. Two quick checks:
- Pull the plug. Turn off your network and run the tool. If it still reads the image, the model is local. If it fails, your image was going to a server.
- Watch the requests. Open your browser's network panel while you OCR something. Local tools make no image upload. You'll see the difference immediately.
Those two tests cut through every privacy claim on every landing page. Real on-device OCR passes both. Textquill is built to pass both, which is the entire point of doing OCR in the extension rather than in the cloud.
The takeaway
"Private" should mean the data didn't move, not that a company pinky-swears to clean up after itself. Screenshot-to-text is one of the most frequent little tasks in a workday and one of the easiest places to leak sensitive data without noticing. Doing it on-device, the way Textquill does, collapses the whole risk into a non-event: fast text out, image staying exactly where it was. If you want the reasoning behind why local models are ready for this, start here. If you just want to stop uploading your screenshots, install the extension and run the offline test above.