A Quenzio build story
Working build, in development
A voice assistant you can cut off
How we built Quenzio, a push-to-talk assistant for Windows 11 that hears and speaks on your own machine and asks before anything risky.
By Doerwise ·
The wish
I want a voice assistant on my own computer. Press a key, say what you want, and hear the answer. Let it open an app or read a file for me.
Why it sounded impossible
Three demands pull in different directions. The listening and the speaking should happen on your own machine. Talking to it should feel like talking, which means you can cut in while it speaks. And an assistant that can run commands on your computer can also delete things you wanted to keep.
Quenzio needed all three in the same loop.
Hearing and speaking on your machine
Every turn runs one pipeline: microphone, speech to text, a language model with tools, the voice, then the speaker. Speech to text and the spoken voice both run locally on your own computer.
You hold the key to talk and release it to send. If nothing usable was heard, Quenzio says "I didn't catch that" once and the turn ends, rather than guessing.
In between, the turn moves through plain stages: listening while you hold, transcribing when you let go, thinking while the model works out whether it needs a tool, then speaking. Ask it to open the calculator and the model calls the tool that opens apps. Ask what is in a text file and it reads the file, then tells you what it found.
Cutting in without breaking it
The language model streams its reply to the voice one sentence at a time, so Quenzio starts speaking before the whole answer is written. The model keeps streaming the next sentences while the speaker plays the first, so a long answer never holds up the start of it.
The voice loop never waits on playback. That one decision is what makes interrupting work. Press the key while Quenzio is talking and it stops, then listens to your new turn straight away. Nothing has to finish first.
Asking before anything risky
Quenzio acts through a small set of tools: open an app, read a file, run a command. Every command is checked against a deny-list of destructive ones before it runs.
When a command matches, such as deleting a folder, Quenzio does not run it. It says out loud what the command would do and asks you to say yes. That yes only counts for the next turn, and the voice is checked before the command goes ahead.
Everything else simply happens. Opening an app or reading a file goes straight through, and the spoken yes is saved for destructive commands, where a mistake would cost you something.
Where it stands
Quenzio is a working build, in development, for Windows 11. It has no window of its own: it lives in a terminal as a push-to-talk loop.
The demo on its product page plays that loop as a console with scripted turns. The audio is simulated, and it never touches your microphone.