private voice ai

shader

speech recognition and voice synthesis
run entirely on your gpu.
your voice never leaves your device.

speech recognition · voxtral · webgpu

voice synthesis · kokoro · webgpu

language model · api · encrypted

preparing

first visit only · ~200 mb · cached forever after

tap to speak

shader

settings

api mode

passkey

enter the shader passkey to access the shared model at no cost.

api key

base url

model

voice

everything local by default. speech recognition and voice synthesis run on your gpu via webgpu. only the language model reply requires a network call. no audio data ever leaves your device.