Airlock, privacy firewall for AI

Redact emails, secrets and client data on-device before your prompt reaches ChatGPT, Claude or Gemini, then restore them.

You paste a lot into ChatGPT, Claude and Gemini. Client names, emails, API keys, patient details, contract terms, things that are not supposed to leave your machine. Airlock catches them before they do. It sits on the page. When your draft contains something sensitive, the Airlock pill lights up. One click swaps each value for a placeholder, so the model still understands your request but never sees who it is about. When the answer comes back, your real values are restored in the reply, on your device. How it stays honest: - Detection runs on your machine. The extension makes no network requests of its own and asks for no network permission. Its only access is to the ChatGPT, Claude and Gemini pages, so it can rewrite your draft before you send it. - The only traffic is your browser talking to those sites as it already does, now with the sensitive parts removed first. - Open your network tab and watch Airlock stay silent. Two layers of detection: - Fast on-device rules for structured data: emails, phone numbers, card numbers (Luhn checked), government IDs, and API keys in many formats. - Pro adds a names model that runs entirely on your device and catches the messy human stuff in plain writing: names, organizations and places. Free forever to redact in the popup and in the page. Pro brings the on-device model that auto catches names, organizations and places, and activates with an offline license key your browser verifies locally. Built by our studio. Airlock raises the cost of a leak to near zero. It is a strong safety layer, not a compliance guarantee.