Voice
Talk to Kawase on iPhone, Android and the web in your language; it answers aloud and shows the same cards as chat. Payments by voice always wait for your tap.
Talking to Kawase
Tap the orb in the middle of the dock (Talk to Kawase in the sidebar on a computer) and speak. Kawase greets you, listens, answers aloud and puts the same cards on screen as chat does.

- The orb moves with the voices
- Listening, thinking or speaking
- What Kawase is saying
- Keep talking in the background
- Type instead
- Mute: nothing is heard or sent
- End
- One conversation. What you said by voice is in your chat thread, and typing picks it up where you left off.
- Captions on both sides: what Kawase heard, and what it is saying.
- A spoken "yes" never approves a payment. A send by voice is prepared on its card and waits for your tap and Face ID (fingerprint or screen lock on Android, an email code on the web).
- Type instead is always one tap away, and mute stops anything being heard or sent.
Keep talking while you use the app
The chevron at the top (or a tap outside, Escape or Back) shrinks the conversation into a black island at the top of whatever page you open: the live orb, what Kawase is doing and its latest words, mute and end. A card Kawase needs you for slides up on that page, and a glow along the screen's edge says voice is on. Tap the island to open the full view again; only End ends the conversation.

- The island: what Kawase is doing and saying, mute and end
- The glow: voice is on
On the web
Voice works in the web app too: allow the microphone when the browser asks. If you refuse it, or there is no microphone, Kawase stays connected, says how to allow it, and typing is right beside it. Kawase's own voice still plays.
Settings → Voice


- Kawase's voice: one of six
- Relaxed · Normal · Brisk
- Kawase speaks first, or waits for you
Tap the speaker beside a voice: Kawase says hello in it right there, without the microphone. Changes apply the next time you start talking.
How it works
The phone or browser joins a LiveKit room over WebRTC, and Kawase's voice worker joins the same room. Every money or account request is handed to the same agent turn as chat (same tools, same limits, same thread); the cards arrive on your screen while the answer is spoken. Before speaking amounts, assets, @handles or the network, the worker checks them against what the tools returned and corrects a mismatch. If the realtime voice can't start, Kawase falls back to a listen → answer → speak pipeline with the same agent behind it. LiveKit doesn't record the conversation.
Voice
What runs today
- talking to Kawase on iPhone and Android in your language: it hears you, answers aloud, and puts the same cards on your screen as chat
- voice in the web app too, through the browser's microphone; if the microphone is refused, Kawase says how to allow it and typing stays beside it
- one conversation: what you said by voice is in the chat thread, and typing picks it up
- a send by voice is ready on its card and waits for your tap (Face ID / fingerprint), even inside the daily allowance
- Settings → Voice: Kawase's voice (six voices), speaking speed (relaxed, normal, brisk) and whether it speaks first
Not yet
- voice inside the desktop app (not checked yet; use the web app in a browser)
- interrupting Kawase mid-sentence reliably on every device (being tuned)