Private replies
Every Petto reply runs on a private NEAR AI model and comes with a receipt. This page says what the receipt proves and what it does not.
Latest attestation
The brain did not answer, so no attestation is shown right now.
What is checked
- [ok]The model runs on GPUs inside a trusted execution environment. The brain checks NVIDIA's GPU attestation with a fresh nonce at least every 30 minutes.
- [ok]Each reply is signed inside the TEE. The brain hashes the exact request and response bytes and checks that the signature covers both.
- [ok]The receipt says which personality (Soul version) and which model made the reply, and links the on-chain fuel charge.
What is not checked
- [--]The Intel TDX CPU quote is not checked in the app.
- [--]The brain itself is not in a TEE. It builds the prompt and can read the pet's memories, which are encrypted at rest.
- [--]A Gateway verified reply was signed by the NEAR AI gateway, not by the model's own key.
- [--]Request and response bytes are dropped after hashing, so a receipt cannot be replayed against the text later.
Voice
Spoken replies use ElevenLabs, a third-party voice service. When a reply is spoken, its text leaves the private model and goes to ElevenLabs to become audio. When you hold to talk, your recording is sent to NEAR AI to be turned into text. The text of every reply is always shown, and each receipt says whether voice was used.
On chain
The pet's fuel, its Soul and every charge are public on NEAR testnet. What you say to your pet is not on chain and is never written to logs. A receipt only becomes public if the owner publishes it.