Medical documents are among the most private things a person owns. This page describes, step by step, what happens to yours when you use MedicDecode, including the parts that are less than ideal.
1. Your file is read on your own device
When you choose a PDF or a photo, it is opened by code running inside your web browser. For PDFs that contain real text we read that text directly. For scans and photos we use optical character recognition (OCR), which also runs in your browser. Both tools are loaded from this website itself, not from an outside service.
The original file is never uploaded. It does not reach our server or anyone else’s. The explainer page also sets a browser security rule (a Content Security Policy) that only allows it to connect back to this site, so that page cannot send your text to any other address and cannot load anyone else’s code.
2. Personal details are removed on your device
Before anything is sent, the text is scanned in your browser for:
- names of patients, doctors and other people;
- dates of birth;
- street and postal addresses;
- phone numbers and email addresses;
- ID, medical record, insurance, policy, account and similar numbers.
Each one is replaced with a label such as [NAME] or [ID NUMBER]. Dates of tests and visits, your age and your sex are kept, because the explanation can depend on them.
3. You check it and decide
You are shown your text with every removal highlighted. You can keep something that was removed by mistake, remove anything that was missed, and tick a box to see the exact text that will be sent. Nothing leaves your device until you click Send this redacted text.
Automatic detection is based on patterns and it does miss things, especially unusual names, handwriting read by OCR, and details buried in free text. Please read the preview. You are the final check.
4. Only the redacted text is sent
When you click send, three things travel to the MedicDecode server: the redacted text, the document type you chose and the reading level. The server passes the redacted text to an AI model together with fixed instructions, receives the explanation and returns it to your browser.
The AI model is reached at (not configured), an OpenAI-compatible endpoint configured by the site operator. Personal identifiers are still removed on your device before anything is sent. At the moment this copy of the site is running in demo mode with no provider key, so nothing is sent to a model and explanations are labelled mock output.
5. Nothing is stored by MedicDecode
- There is no database of documents or explanations. We do not keep either.
- The server writes one log line per explanation with the number of tokens used, the estimated cost, the time taken, the document type and the reading level. It does not log any document text.
- To limit abuse, the server keeps a short-lived count of requests per IP address in memory. It is not written to disk and is cleared when the server restarts.
- The explanation lives only in your browser tab. Closing or reloading the page removes it. Use the print button if you want to keep a copy.
- The explainer page sets no cookies and has no advertising or analytics scripts. Adverts, where the site shows them, are only on reading pages that never hold your document.
What you should weigh up
- The medical content of your document (results, findings, medicines, charges) is sent to the AI provider. Only the identifying details are removed.
- Redaction lowers the risk that the content can be linked to you. It cannot remove it completely: a very rare condition or an unusual combination of details can itself be identifying.
- If you are not comfortable with that, do not send the document. You can still use the first two steps to read the text from a scan, and close the page without sending.
The formal version of this information is in the privacy policy. If something here is unclear, get in touch.