// browser extension · fully local
Stop personal data before it reaches an AI chat.
Mr. Rao Plus watches what you are about to send to ChatGPT, Claude, Copilot, Gemini, Grok, DeepSeek or Perplexity.
If it finds an IBAN, tax ID, card number or name, it removes it from the box — or stops the send and explains.
No network calls to us. One permission: storage.
Google Chrome
Microsoft Edge
Firefox// 01 · not a clone
Not Mr. Rao under another name.
Mr. Rao (desktop) works on documents. Mr. Rao Plus works on the chat box: typed text, attachments you pick, outbound request bodies. Same recognition engine, same author, two jobs.
// DLP · the gesture
DLP for AI chats, with no extra server.
Shadow AI is a colleague pasting a case file into ChatGPT. Network DLP sees the packet late. An AI proxy sees the prompt — on their servers. Plus sits in the box and on the outbound body, on the device. It does not replace enterprise DLP. It covers the hole that DLP does not see. Listed chats only: that is the price of the promise.
What it is not
Not network DLP. It does not see mail, USB sticks, or SaaS. No console of ours, no extra DPA toward us.
What it is
The gesture: paste, attach, Enter. Three outcomes, not a traffic light. If it cannot look, it is not clean. If the hooks fall off, it says so.
// 02 · how it decides
Three outcomes, not a generic red light.
Stop
IBAN, tax IDs, cards, emails, phones, names, secrets… The message does not leave until you fix it.
Warn
Amounts, dates, web addresses: it sends, and tells you. Off by default — a warning that always arrives is one you learn to dismiss. Switch them on in the settings.
Could not look
Opaque attachment, image-only PDF: never says “clean”. Blocks and says so.
// 03 · where it runs
Only listed AI chats.
No <all_urls>. Anywhere else the extension does not exist.
// 04 · FAQ
Questions that matter when you actually use it.
If I accept the edits and then type a tax ID, does it block again?
Yes if you only corrected or closed with “Understood”: every send is checked from scratch. A new tax ID is blocked again.
There is no blanket “personal data is fine in this chat” permission.
What does “Send it anyway, once” do?
A one-shot override, same on the box and on the network (from 0.1.15):
- the next send that would block goes through once;
- the override is consumed;
- a clean message does not eat it;
- the second dirty send is stopped again.
It is not a two-minute window where protection is off.
“This is not personal data”?
Adds that value to the “never touch” list. Only that value. Another tax ID or IBAN is still stopped. Permanent until you remove it in settings — and the UI says so.
Does Ctrl+Z undo redaction in the box?
Often yes on contenteditable chats: rewrite uses insertText, which browsers put on the undo stack. If you restore and resend, protection checks again (unless a one-shot override is still unused).
Are attachments checked?
Yes (recent versions): when you pick a file, Plus tries extractable text. If it cannot (scan, image, audio) it does not say clean — third outcome and block. No OCR. No upload to our servers.
Does it send data to valor-cyber?
No. No host_permissions to us. Offline licence verification when needed. See the privacy notice.
Same engine as desktop Mr. Rao?
Yes: TypeScript port of the same engine, with a conformity corpus vs the Python original. Plus adds browser hooks, the box, attachments and override — not a second invented recognizer.
Free? Paid later?
0.x is free without a key. If paid versions come: protection keeps working after expiry; only what the extension says changes. First-install date is already stored locally.
Declared limits
- Automatic recognition errs both ways.
- Names: mainly Italian lists; optional offline EN extra.
- If a site changes host, an update is needed — the watchdog cannot watch a page it is not injected into.
// 05 · install