BRAINreply intelligence
Gemma 4 E2B
Browser-optimized LiteRT-LM intelligence.
GNM HEAD
v3.0
MEAN IDENTITY
67 MORPHS
INTERACTIVE DEMO
ON DEVICE
02 / LIVE AVATAR
Meet your local talking avatar.
Speak naturally and watch a local agent listen, think, answer, and articulate in real time. Choose a brain and voice below—the full conversation runs in your browser.
DEMO STATUS
ready to load
MICROPHONE INPUT
MICROPHONE PAUSED
Input is off. Start listening when you’re ready.
- 01LISTENMoonshine Tiny q8
- 02HEARwaiting
- 03THINKwaiting
- 04SPEAKwaiting
- FIRST TOKEN
- —
- FIRST CLAUSE
- —
- TTS SYNTH
- —
- TURN → AUDIO
- —
YOU
Start the demo, then speak naturally.
LOCAL THOUGHT visible here · never spoken
AVATAR
Your avatar’s reply will appear here. First run downloads the selected local models once.
LISTEN Moonshine Tiny34 MB
THINK Gemma 4 E2B2.0 GB
SPEAK Kokoro Timestamped fp32326 MB
Local model manager checking browser…
VOICEspeech synthesis
Kokoro Timestamped fp32
WebGPU voice synthesis with native phone durations.
LIP SYNCGNM timing policy
Acoustic refinement
IPA windows refined by local energy, voicing, and transient cues.
MIC INPUTspeech recognition
Moonshine Tiny q8
Low-latency English speech recognition for local conversation.
Models load only when you ask.
Model inference stays on this device. Starting the demo confirms each model’s one-time download; switching a model unloads its previous runtime first.
- WebGPU
- checking
- Cache
- checking
- Local loop
- zero cloud calls
Try any written phrase ElevenLabs voice + timed lip sync
Add your own API key to generate a one-off phrase. The local conversation demo above does not need this key.
API key required
not configured
Stored in this origin’s local storage and sent directly to ElevenLabs when you generate. Site scripts can read it—use a restricted, low-quota key and never save one on a shared device.
Ready for a phrase
no audio loaded
ACTIVE GESTURE
articulatory rest
JAW
LIPS
TONGUE
IPA / DETERMINISTIC PHONE WINDOWS
awaiting speech
How speech becomes motion
The outer lips and mouth interior belong to the same native GNM mesh. Teeth and tongue are the model’s own internal anatomy; no detached facial shell or replacement mouth is used.