Nixi

A desktop voice assistant for Linux. Trigger Nixi, ask a question or give a command, and it handles it — change the volume, set a wallpaper, take a screenshot, control media, or answer general queries using Gemini.

What It Does

• System control — volume, brightness, mute, media playback, screenshots, wallpaper changes, app launcher.
• General queries — ask anything. Nixi uses Gemini and can search the web for live information like weather, news, and prices.
• Vision — say "open your eyes", "where should I click", or "why is this red" and Nixi asks permission to screenshot your screen, then describes what it sees. Approval is via desktop notification with a configurable timeout.
• Voice interaction — manual trigger-to-record mechanism, speech-to-text, and text-to-speech. No constant background listening/wake-word overhead.

Installation & Setup

Option 1: Global Installation (Recommended)

Install Nixi globally on your system using pipx:

pipx install nixii

This will automatically expose the commands globally. Run any command once to automatically initialize your configuration directory:

nixi-server

Now, configure your API keys by editing ~/.config/nixi/.env:

GOOGLE_CLOUD_API_KEY=your-vertex-express-mode-key
SARVAM_API_KEY=your-sarvam-key

Option 2: Local Development Setup

If you'd like to run or develop Nixi locally, install uv and sync dependencies:

uv sync

Run any command once (e.g. uv run nixi-server) to auto-generate your config, then add your API keys to ~/.config/nixi/.env.

Run

Note: If you are using the local development setup, prefix all of the following commands with uv run (e.g., uv run nixi-server).

First, discover your desktop environment (keybinds, scripts, wallpapers):

nixi-server --discover

Then start the server:

nixi-server

Next, start the voice daemon (which runs as a background process listening for manual trigger events):

nixi-voice

To interact with Nixi, trigger the recording. You can bind keys or desktop shortcuts to run these commands:

• Trigger Recording (Toggle): Start or stop recording your query.

  nixi-voice trigger
  

• New Session: Start a fresh conversation session and trigger recording.

  nixi-voice new
  

Run the command once to start recording, speak your request, and run it again to stop and process the audio!

Configuration

Nixi resolves its config in this order:

1. --config <path> CLI flag
2. $NIXI_CONFIG environment variable
3. ~/.config/nixi/nixi.toml (user config)
4. example_config/nixi.toml (repo default)
5. Code defaults (no file needed)

The system prompt lives in code. Set agent_name in your TOML to change the persona — the prompt template uses it automatically.

Edit your TOML to configure audio thresholds, STT, TTS, model settings, vision, and local actions.