NewTyger · full-duplex voice agent
IT LISTENS WHILE IT TALKS.
Meet Tyger, a full-duplex voice agent. Cut in anytime: it stops, listens and replies in about 250 ms. No ASR → LLM → TTS pipeline to stitch together.
The messy parts of talking don’t trip it up.
Cut in mid-sentence and it stops to take your correction.
Tyger…they have a table at seven,
YouMake it eight.
Yields instantly
It says “mm-hmm” while you talk, without taking the turn.
YouHey, I need a table for two tonight,
Tygermm-hmm
Pause to think and it waits instead of jumping in.
YouI’d like the… um…
Youthe salmon.
Waited for you
It calls your functions and actually places the order.
order_pizza(size="large", toppings=["mushrooms"])
Order placed
Use it on our API, or run it yourself
Run it yourself
Open weights, commercial use included. Your audio stays on your GPUs.
Just need the voice?
Gepard, our streaming text-to-speech: under 100 ms, 15 languages, ~22 hours of audio for $15.