Many people thought it was impossible to make a worse voice assistant than Siri. I accept that challenge.

STT is handled by a hard-coded faster-whisper, probably moving to Parakeet for Mac (since faster-whisper uses CUDA).

TTS is currently handled by Kokoro and not satisfactory, so I've made it off by default. It can be turned on in the settings panel.

Repo breakdown

tbc

Installing

With the script

Run

.\install.bat          # CPU
.\install.bat gpu      # + NVIDIA CUDA speech-to-text

Manually

You'd need Python 3.12+. I recommended that you use an isolated environment (.venv) so installs do not touch the system Python.

First, enable Long Paths, because PySide6 (which drives the UI) nests QML module trees exceeding 260 characters. The install currently silently fails without it. After install, run python -m frontend to check for this at startup.

Then, create a .venv and install CUDA for NVIDIA GPUS:

python3.12 -m venv .venv          # Windows: py -3.12 -m venv .venv (optional)
source .venv/bin/activate         # Windows: .venv\Scripts\activate (if .venv is used)
pip install -e .
pip install -e ".[gpu-cuda]"      # optional: NVIDIA GPU speech-to-text (~28x faster)

Then, to start, start the assistant with:

source .venv/bin/activate         # Windows: .venv\Scripts\activate (optional)
python run.py                     # start BOTH — daemon + Teleprompter — in one window; Ctrl-C stops both

Some custom checker commands:

tbc