Talk, do not dictate
Voice activity detection understands when your turn ends and keeps the conversation moving.
DSH
KITT.
A voice layer for DeepSeek Harness. Talk naturally, hear the answer while it arrives and keep the conversation with you anywhere on the desktop.
Your coding agent
leaves the tab.
DSH KITT is built for conversation, not dictation. It detects when you stop speaking, sends the message, waits for the agent and reads the reply sentence by sentence — without breaking your flow.
01—04 Built to be used
Always within reach
A compact overlay stays above your work so you can talk from the editor or any other application.
Choose every signal
Select microphone, audio output, language and voice, with neural, system and local options.
Operate from anywhere
Global shortcuts keep listening, mute and playback controls accessible beyond the browser.
1.0 Released 5 September 2026
A conversation that holds
Replies that used to stall halfway, or leave the microphone deaf after an interruption, now finish and hand the turn back every time. Every fix in this release came from daily use.
Consistent from the first word
Engine, voice and pace are decided once per reply and kept to the end. If the chosen engine fails, the reply continues with the built-in voice and the status line says so.
Keys go where you are
With the harness open in more than one place, global shortcuts reach only the page that is using the voice, or the one in front of you. The others stay quiet.
One control for KITT mode
The brand mark switches hands-free conversation on and off and shows the state in colour: blue while waiting, green while listening, red while speaking.
Available now
One command
and it talks.
Free and open source, MIT licensed. Listed in the DeepSeek Harness plugin catalogue.
dsh plugin --profile web add dsh-kitt-voice