Aetox harness
v1.11.19Aetox 1.11.19 is out. See what changed

Make it yours

Sound: speak instead of typing, and listen back

Speech-to-text runs on your machine by default, and the reading voice is Edge, free — every row that sends sound out says so.

Two directions: the microphone in the composer to speak instead of typing, and a listen button under a reply to read it aloud — plus an audio_transcribe tool the agent uses on an audio file.

In the app
Settings › Sound — pick the provider, the model and the voice, with a speed slider for reading
Speech-to-text default
on this machine (faster-whisper) — it needs Python installed · the other local choice is whisper.cpp
Reading default
Microsoft Edge — free, no key · locally it is the Windows voices or Piper
Needs a key
OpenAI · Groq · Gemini · Mistral · Meta · Xiaomi · ElevenLabs — choose one only when you already have the key
Where the models live
%APPDATA%\Aetox\models (Piper's voices in %APPDATA%\Aetox\tools\piper)
Where the choice is kept
%APPDATA%\Aetox\model-preference.json

Run on your own machine

Free, no key · prompt never leaves the machine

ollamalmstudiocomfyui (pictures)stable diffusion webui (pictures)

Use a provider you trust

Enter a key for one you already have, or log in with GitHub Copilot · ChatGPT · OpenRouter · talk straight from your machine to them, not through us.

github copilotkiloanthropicopenaigeminideepseekkimiminimaxopenroutermistralqwengroqzaixaithaillmmodelscopenvidiaollama cloudchatgpt (codex)opencode zenopencode gometaxiaomi mimoxiaomi mimo token plan

Click a name to see each provider's details · plus any OpenAI-compatible endpoint of your own · switch at any time, no lock-in