Eyes widen with loudness, blink on an irregular human cadence, and follow your cursor. Arrow keys aim the gaze, C re-centres it, and Bubble.lookLeft() / lookRight() drive it from code.
Uses the browser's built-in speech engine. Browsers don't expose that audio to the visualiser, so the orb follows a syllable envelope estimated from the text — for true waveform reactivity use Demo voice, Mic, Audio file, or Live AI.
Speaks to the OpenAI Realtime API over WebRTC straight from this page. The orb analyses the model's actual returned audio. The key stays in memory only — nothing is stored or sent anywhere else. Browser-side keys are fine for local play, not for anything public.
Any voice stack works — drive it from the console or your code:
Orb.connectStream(mediaStream)
Orb.connectNode(node, ctx)
Orb.connectElement(audioEl)
Orb.setLevel(0…1)
Orb.setState('speaking')
Orb.pulse(0.8)
drag · rotate | scroll · zoom | space demo | M mic | S settings | F fullscreen | H hide UI | E eyes | ← → look | C centre | Esc stop