Kokoro Voice

Reads the assistant's answers aloud in a natural voice, made on your own computer's graphics card. Nothing you hear is sent anywhere.

What you can do

Ask the assistant something by voice and hear the answer. Each sentence is spoken as soon as it is ready, code is skipped, and pressing the microphone stops the reading.

Getting started

Dictate a message in the assistant. The first answer is read by your system's voice while the neural voice downloads once, about 326 MB; later answers use Kokoro.

Capabilities

Capability	What it does
speech.speak	Reads a text aloud, one sentence at a time.
speech.stop	Stops the reading that is playing.
speech.voice.prepare	Downloads the voice once and warms the graphics card.
speech.voice.status	Reports the voices on offer and whether this machine can run them. Reads only.

None of them is offered to an agent as a tool: the speaker is yours.

Requirements

• Desktop (Electron) app on a machine with WebGPU.
• English voices: Heart, Bella and Michael (US), Emma and George (UK).