Questions, answered.
The short versions. Every answer links to the longer one in the docs.
Is it actually private, or is that just marketing?
Privacy here is a property of the architecture, not a promise in a policy. You sign in with a passkey that lives on your device, so there is nothing to look you up in. Prompts are encrypted in your browser so only the operator running your model can read them, and your history stays on your device. Where privacy has limits, we publish them. How privacy works →
What can an operator see?
The operator serving your request decrypts it to run the model, so it sees the prompt and the reply. It cannot see who you are: behind a request there is only a payment address, and requests reach it routed through a peer, not from you directly. The full threat model →
What happens if I lose my passkey?
There is no “forgot password”, because nothing exists anywhere else to reset. Your passkey manager syncs it across devices, and Settings → Recovery gives you a 24-word phrase that can move your balance to a fresh passkey. Without either, a lost passkey means a lost balance, so keep only what you will use on hand. How recovery works →
Why does a privacy tool use a blockchain?
Only payment touches it; prompts never do. Settling each message on a public ledger is what lets us bill you without knowing who you are. The trade-off is that payments are public: an observer sees an address paying amounts, not what was asked and not whose address it is. Why a chain, in plain terms →
How do I add funds?
Top up from inside the app by card or debit, or send crypto from a wallet you already have. Either way it lands as dollars, and nothing is billed to you later. The balance is yours, and you can withdraw it. How billing works →
Is the client open source?
Not yet. It is still changing fast, and publishing a security-critical codebase mid-churn helps nobody relying on the audit. What the client sends is inspectable from your own network tab, and we document exactly what each party can see. Our open-source record →
The rest is in the docs.
Threat model, protocol, running an operator. Or skip the reading and see for yourself.