Where the future is rough and running
An early look at the experiments brewing in our research lab. Some of these ship as products. Some teach us what not to build. All of them are real.
What we're trying
Polyphonic voice
Generate two distinct speakers in a single stream, with natural turn-taking and overlap. Currently in closed preview.
Request preview accessReal-time avatar emotion
Drive an avatar's facial expression from the emotional state of the speech in real time, frame-by-frame.
Request preview accessSub-100ms voice latency
A new inference path that gets time-to-first-audio under 100ms for true real-time conversation.
Request preview accessOn-device agent memory
A compact episodic-memory module that runs entirely on-device, with no data leaving the phone.
Request preview accessThe problems we keep returning to
Speech synthesis
Expressive prosody, emotion control, and sub-100ms latency.
Avatar & vision
Real-time lip-sync, gaze, and expression from audio alone.
Conversational agents
Memory, tool use, and reasoning that stays in character.
Safety & provenance
Consent, watermarking, and misuse detection built in.
What 'lab' means
Things may break
These are research previews. APIs can change, and outputs can be rough.
We share what we learn
Every experiment ships with a write-up of what worked and what didn't.
Your feedback shapes the roadmap
The most-requested previews graduate to production features.
Want to play in the lab?
Preview access is limited and granted to active accounts first.
SOC 2 Type II · GDPR · HIPAA-ready · No vendor lock-in