Private-AI Chat (portal)

Chat is the CloakAPI front door for people who just want to talk to a model privately — no code, no install. It runs in your browser at app.cloakapi.io/chat and puts a tokenisation layer between you and Claude, GPT, Gemini, Grok or DeepSeek. Personal data in your prompt (and in files you attach) is tokenised in your browser before the request is relayed through the blind gateway to the provider; the model’s reply is turned back into the real values locally, so the provider only ever sees <<CLOAK:CONTACT_EM:…>>, never alice@meridian.example.

Chat is one of the three ways to use CloakAPI (API / SDK / proxy · Chat · browser extension). If you write code, start with the Quickstart; if you want the same privacy layer on ChatGPT or Claude’s own site, use the browser extension.

Get started

Create an account at app.cloakapi.io/signup. Sign-up grants a small trial credit so you can send your first messages right away.
Open Chat at app.cloakapi.io/chat and pick a model. You can use CloakAPI’s pooled provider keys or bring your own (BYOK) — set that under Integration → BYOK in the portal.
Type normally. Names, emails, phone numbers, cards, national IDs and other structured PII are tokenised on your device as the message is sent. The reply is detokenised back for you.
Keep a receipt. Every response carries an ECDSA-P-256 signed receipt you (or an auditor) can verify later at app.cloakapi.io/receipt-verifier.

Good to know

Paid, key-backed, fail-closed. Tokenisation requires an account with balance. If Chat cannot tokenise a message — paused, no key, or zero balance — it blocks the send rather than let raw text leak. Your original text never leaves the browser in that case.
Pricing is usage-based, USD only. A flat markup on top of the underlying model cost — no subscriptions, no seats. See pricing.
Try tokenisation with no account in the in-browser tokenisation playground.
Want the fullest on-device coverage? Pair Chat with the Windows desktop app — a power-up that adds the native tokeniser and complete-local models.

For exactly how the tokenisation and blind-relay model works, read How privacy works.