Voice desk: viora desktop

A hands-free assistant on your own computer. Say “viora”, ask anything in English or Swahili, and hear the answer while it's still being written. It thinks with the AI you already use, and can take tasks that it finishes in the background.

Install

macOS and Linux:

curl -fsSL https://viora.depriver-tech.workers.dev/install.sh | sh

Windows (PowerShell):

powershell -ExecutionPolicy ByPass -c "irm https://viora.depriver-tech.workers.dev/install.ps1 | iex"

Or with Python 3.10+: uv tool install "viora-ai[desktop]" (or pipx install "viora-ai[desktop]"). Then:

viora setup           # choose the AI, ears, voice and wake word; plays a test
viora app             # viora in its own window
viora desktop         # or the full-screen terminal UI
viora app --shortcut  # add viora to your applications menu
viora doctor          # check microphone, speaker, voice, ears and AI

Two ways to see it

viora app opens viora in its own window: an animated orb that shows what it's doing (waiting for its name, listening, hearing you, thinking, working, speaking), the conversation as it's written, your background tasks, a text box, and buttons to stop it, turn the microphone off or start over. It uses Chrome, Edge, Brave or Chromium as an app window when one is installed, or your default browser. Closing the window ends viora.

viora desktop runs the same assistant as a full-screen terminal UI: the logo and your settings at the top, the conversation in the middle, an animated status bar with a live microphone meter, and an input line. Esc stops viora talking, Enter sends, Ctrl+C quits. viora desktop --plain prints simple lines instead, for logs and basic terminals.

Choose its brain

viora runs the official command-line tool you have installed and signed in to, so it uses your existing subscription. Agent brains can search the web and act on your computer; API brains are fast and conversation-only.

With Claude Code, one process stays open across questions, so only the first answer waits for it to start. Conversation turns use low effort for speed; tasks use full effort.

What it may do

Ears, voice and wake word

• Ears (speech to text): Groq Whisper (free tier, fast), Cloudflare Whisper (free daily allowance), OpenAI, or local faster-whisper, offline (pip install "viora-ai[local]").
• Voice: Microsoft neural voices (edge, free, no key) are the default, with natural English and Swahili. Also ElevenLabs, Azure, Cloudflare Aura, OpenAI or system. Pick female or male, or a name: see Voices.
• Wake word: say it before a question (“viora, …”). For 20 seconds after an answer, follow-ups don't need it. Set wake: "" to always listen.
• Interrupting: while viora speaks, say its name (“viora”), or “stop” / “acha”, and it stops at once and listens. “viora, what about tomorrow?” stops it and asks the new question. It recognises its own voice coming back through the speakers and ignores it. You can also press Esc or the Stop button. With headphones, set barge_in: true and any speech interrupts it.

Things to say

You can also type at any time; /stop, /reset, /mute, /help and /quit work in the terminal. While the AI uses a tool, viora says “one moment” / “subiri kidogo” so it never goes quiet, and soft sounds tell you when it heard you.

Settings file

viora setup writes ~/.config/viora/desktop.yaml (Windows: %APPDATA%\viora\desktop.yaml) and keeps keys in a .env file next to it, readable only by you. Every setting can also be set with an environment variable such as VIORA_DESKTOP_BRAIN=opencode, or for one run with viora desktop --brain opencode --voice elevenlabs.

name: viora
user: Priver
brain: claude
model: sonnet
actions: safe
workdir: ~/Projects
ears: groq
voice: edge
voice_en: andrew        # or female / male
voice_sw: daudi
wake: viora
language: auto          # auto | en | sw
barge_in: false
sounds: true