v0.21.0

Neko's own voice — at the scene, not through a robot.

Highlights

• The camera coach speaks with Neko's real voice now. It used to read cues through the phone's
  built-in text-to-speech. Tap Giọng Neko on the camera page and the phone holds an actual
  GPT-Live session: audio flows phone ↔ OpenAI directly (the terminal never carries it), the voice is
  Neko's warm, natively-Vietnamese one, you can interrupt it, and you can talk back — ask for
  another pose idea, tell it to wait — while the camera keeps feeding it what it sees.
• Spoken turns from the phone land in the terminal transcript as (voice · phone) and join the
  session history, so the conversation continues after you put the phone down.
• Device text-to-speech stays as a fallback for the cue text and switches itself off the moment the
  real voice is live.

Privacy

The camera page now asks for the microphone as well — microphone=(self) on that page only; the
main client page keeps its full lockdown. Camera and microphone both start on an explicit tap and are
released when you leave the tab or lock the phone. The SDP handshake is end-to-end sealed like every
other relay payload.

Install / update

neko update — or:

irm https://neko.holilihu.online/install.ps1 | iex

curl -fsSL https://neko.holilihu.online/install.sh | sh

Full detail: CHANGELOG 0.21.0