Thinking UI

Voice ModePRO

A full-screen voice conversation: a reactive orb, live word-by-word captions and call controls.

Code

11 files · 1,900 lines of TypeScriptPRO

Voice Mode is part of Pro

Get its source, the one-paste agent prompt and all 113 components, for a one-time $99.

Motion notes

  • One orb carries the whole conversation: it listens, thinks and speaks through the orb's own states, swells slightly while the assistant talks and draws in while it thinks. Pick the orb with `orb`; feed `level` from your audio graph to make it follow the real voice.
  • Captions reveal word by word even when the transcript arrives in chunks, catching up to whatever text has come in. Only three lines stay: the newest at full size, older ones step down, dim and drift up before they leave.
  • Controls are labelled under each button like a call screen. Space mutes, Esc interrupts while the assistant is thinking or speaking, C toggles captions; hotkeys ignore text fields and inert previews.
  • The state label and its hint swap with a short vertical slide so a change of turn is noticed without looking at the orb. With reduced motion, the orbs hold still frames and captions appear whole.

Usage

Example
<div className="h-dvh">
    <VoiceMode
      state={state}           // from your realtime session
      level={level}           // AnalyserNode RMS, 0–1
      captions={transcript}   // [{ role: "user", text }, { role: "assistant", text }]
      muted={muted}
      onMuteChange={setMuted}
      onInterrupt={() => session.cancelResponse()}
      onEnd={() => session.close()}
    />
  </div>

  // Without a speech-to-speech API, wire it to the same /api/chat route as
  // Chat App: transcribe the mic, send() the text with useChat, set state to
  // "thinking" while status is "submitted", stream the reply's text into an
  // assistant caption, and speak it with your TTS of choice ("speaking").

Props

PropTypeDefaultDescription
state"idle" | "listening" | "thinking" | "speaking"—Whose turn it is.
levelnumber—Live audio level 0–1: the mic while listening, playback while speaking. Omit to simulate.
captions{ id?, role: "user" | "assistant", text }[][]The transcript. Grow the last item's text as it streams.
orb"particles" | "voice" | "glass" | "halo" | "blob""particles"Which orb sits in the middle.
orbSizenumber280Orb size in px.
muted / defaultMuted / onMuteChangeboolean / boolean / (muted) => void—Microphone mute, controlled or not.
onInterrupt() => void—Cut the assistant off. Enabled while thinking or speaking.
onEnd() => void—End the call.
showCaptions / defaultShowCaptions / onShowCaptionsChangeboolean / boolean / (show) => void— / true / —Captions visibility.
startedAtnumber—Call start (ms since epoch). Without it the timer counts from mount.
wordMsnumber90Milliseconds between revealed caption words.
hotkeysbooleantrueSpace, Esc and C shortcuts.
titleReactNode"Voice"Shown top left.
classNamestring—Classes for the root. Give it a height.