DeviceNexus Demos
These demos are running on a single NVIDIA DGX Spark.
Live
PersonaPlex
Full-duplex speech-to-speech conversation with a 7B parameter model. FP8 quantized, <80ms latency — no cloud inference.
By NVIDIA, based on the Moshi architecture. Running on NVIDIA DGX Spark.
Coming Soon
More Demos
Additional on-device AI demos are in development.
Choose a scenario
Pick who you'd like to talk to. Each persona has a unique voice and personality.
Before we start
Connecting...
Position: —
Estimated wait: —
2:00 remaining
Microphone ready
NVIDIA GB10 · Waiting for stats...
Session complete
Thanks for trying the demo.
or
Want to stay in the loop?
We'll only reach out for major updates. No spam.
Demo is busy
All sessions are currently in use. Please try again in a moment.
Something went wrong
An unexpected error occurred.