What PII does your text leak to the LLM?
Paste a prompt, log line or API payload — see exactly what OpenAI, Anthropic or Mistral would receive, and what's left once the personal data is masked.
Want this automatically in ChatGPT & Claude? Browser extension, 7 days free →
Why this matters
Every prompt you send to a cloud LLM — including the customer record, the log line, the support transcript — arrives at the provider (OpenAI, Anthropic, Google) in cleartext. A no-training flag doesn't change that: the data is still exposed to a third country and a third party.
Under the GDPR that's a transfer of personal data you may not have a basis for. The fix isn't "stop using AI" — it's masking the PII before the request leaves your service, and un-masking the response.
That's what Saklam is: the same detection as here, only automatic on send in ChatGPT, Claude, Gemini & co. As a browser extension, installed in 2 minutes, 7 days free, and with a larger detection model than this tool. For teams and IT as a Docker gateway in your own infrastructure.
Try the browser extension →Chat with a real AI model — masked? That now runs on the hosted Saklam Bridge: OpenAI-compatible endpoint https://bridge.saklam.com/v1, your Saklam licence key is the API key, €10 of model credit included. No Docker, no provider key of your own — in Langdock, Open WebUI, LangChain, Cursor or via curl.
Test data only: the endpoint runs on our server, so the masking happens there. You get the key with the trial.
Rather run it in your own infrastructure right away? Saklam Bridge (Docker) →
How this page works
Detection runs entirely in your browser via Saklam's PII engine (300+ validated patterns across EU-27 + EEA + UK, plus a multilingual NER model for names). Your text never leaves your browser, so there are no logs of it. The model is loaded once from saklam.com (noez GmbH, EU) and cached locally — offline after first load.
