核心要点 (BLUF)： Voice chat with an AI companion right in your browser: press the microphone and speak — the companion answers out loud and remembers what you talked about for next time. Everything runs on your device: recordings and conversation never leave it, no tokens are charged.
The first time you open the chat, the browser downloads the companion models (about 2 GB) — a one-time download that is cached afterwards. Press the microphone and say something: the companion hears you, answers in a live voice and shows its replies as text at the same time.
The companion behaves like a person: it greets you, keeps the conversation going, remembers your name and recalls what you discussed earlier. Answers stream in real time, like in a live conversation.
Every turn is stored in the chat memory inside your browser. When answering, the companion retrieves the memories relevant to the current topic: tell it your name, close the page and come back — you will be recognised.
You can clear the memory at any moment, and the companion starts fresh.
Speech recognition, the companion and the speech synthesis all run locally on your device: no microphone recordings and no messages are sent to our server. No tokens are charged — the chat is free and unlimited.
You need a WebGPU-capable GPU (recent Chrome or Edge) and a one-time download of about 2 GB of models. Phones and browsers without WebGPU are not supported yet.
Yes. No tokens are charged and there is no paid API behind it.
Nowhere. Recognition, the companion and the voice are computed in your browser, and the chat memory is stored only on your device.
The companion models (about 2 GB) are downloaded once and cached; the next conversations start quickly.
It is a small model running right on your device, so it can make mistakes or invent details now and then. Quality keeps improving with updates.
Open the chat settings and choose to clear the memory — the companion forgets the whole conversation context.