Skip to content
OCR2text

Privacy

Your documents don’t need to be uploaded to our server

That is not a policy promise you have to take on trust — it is how the application is built. Recognition happens inside your browser, so there is no upload to secure, no storage to breach and no retention period to argue about.

Last updated 8 September 2026. Plain English, because a privacy claim you cannot check is not worth much.

Never leaves your device

  • Your image or PDF, at any point
  • The recognised text
  • File names
  • Word boxes or confidence scores
  • Anything in a URL, share link or query string
  • Anything attached to an error report

What is downloaded to you

  • The application itself — HTML, JavaScript and CSS
  • The OCR engine's WebAssembly core, served from this site
  • The recognition model for the language you pick, cached after the first use
  • The PDF.js worker, when you open a PDF

Check it yourself

Open your browser’s developer tools, go to the Network panel, and run a recognition. You will see requests for the application code, the WebAssembly core and a language model — and no request carrying your image. Turn your network off entirely after the first run and OCR2text keeps working. That is the honest test, and it is the one we hold ourselves to before each release.

About the model download

Being precise matters here. The recognition engine needs a language model, which is a data file. English is served from this site directly. Other languages are fetched once from the Tesseract project’s model CDN, then cached on your device. That request says which language you asked for — it carries nothing about your document, because your document is never involved in it.

No accounts, no history

There is no sign-in, no email address and no user record. Pages you add live in the tab’s memory for as long as the tab is open. Close it, and they are gone. There is no server-side scan history, because there is no server that ever saw the scan.

Analytics and error reporting

If usage measurement is added, it will be limited to counts of events like “OCR started” or “PDF mode used”, with no document content of any kind — no image, no text, no file name, no word boxes. Error reports are sanitised on the same principle: a stack trace may tell us something broke, never what you were reading.

Exports and sharing

Every export — TXT, Markdown, DOCX, CSV, JSON, searchable PDF — is generated in your browser and saved straight to your device. OCR2text deliberately does not offer cloud-hosted share links. Use your own device’s share sheet if you want to send a file on; privacy is worth more than a convenient URL.

Camera and clipboard

Camera permission is only requested at the moment you choose the camera, never on page load, and the video stream is stopped as soon as you close it. Clipboard access is only used for the image you deliberately paste.

Who runs this

OCR2text is built by Techvora. Questions about any of this are welcome. If you want the detail of how the pieces fit together, the how it works page goes through it.

Try it on something you would never upload

That is rather the point. Open the workspace and watch the network panel while you do it.

Open OCR2text