◈ Aether

Talk to your computer. It listens, and it acts.

Aether is a self hosted voice assistant for your Linux desktop. Speak or type from your phone, laptop, or tablet, and it runs things on your machine and talks back. You choose the AI model behind it, and everything stays on your computer unless you decide otherwise.

It answers in the calm, dry voice of a butler who is impressed by nothing, so even the small stuff feels like a small luxury.

---

What it can do

Just say what you want. A few examples:

• "Lock the screen", or "unlock it"
• "Play some jazz on YouTube", "turn the video down", "skip this", "stop"
• "Read my notifications"
• "Set the volume to 20", "mute my mic"
• "Dim the screen to 30"
• "Find my budget spreadsheet"
• "Open github.com", or "open my downloads folder"
• "Copy this to the clipboard", or "what's on my clipboard?"
• "Take a photo with the webcam"
• "How much RAM is free?", "what's my battery?", "am I on wifi?"
• "Open Chrome", "how many windows are open?", "close this tab"
• "What's the weather? Do I need a jacket?"

This is just a taste. Aether is a real agent, not a fixed list of commands. When there is no ready made action for something, it looks around your machine, finds the right tool, and runs it, so it can do far more than the examples above. It also knows which tools are installed here, so if something needs setting up it tells you plainly instead of failing. When it is unsure it asks you a quick question in the web page, and anything risky waits for your approval first. Full list in docs/CAPABILITIES.md.

Install

From inside your desktop session, open a terminal and run:

git clone https://github.com/isaka-james/aether && cd aether
bash scripts/install.sh

The installer asks a few simple questions (your login, which AI model, what to set up), then starts everything and shows you the link to open. It also sets Aether to start every time you log in, so it is always there. The first run downloads a few things and takes a few minutes.

To remove it later: bash scripts/uninstall.sh.

Updating

To get the latest version, from the project folder run:

bash scripts/update.sh

It pulls the new code, keeps your settings and data, rebuilds the backend, and restarts the assistant. Your downloaded voice models, history, and favourites stay where they are. The database updates its own structure automatically, so an update never breaks your data. If a new setting was added, it is appended to your .env with a sensible default so you can tweak it.

Use it

Open the link the installer printed and sign in:

• On this computer: http://localhost:8473 (voice works here)
• Another device on your wifi: http://your-pc-ip:8473 (typing works here)
• From anywhere: run ngrok http 8473 and open the https link it gives you. This is the easy way to reach your PC from outside your home, with no router setup.

Tap the orb and talk, or type in the box.

Talking by voice on a phone

Phones (especially iPhones and iPads) only allow the microphone on secure https pages. So for voice on a phone, open the ngrok https link, not the plain http address. Typing always works, even without https, if you would rather just chat.

Tip: in your browser menu, choose Install or Add to Home Screen. Aether then opens like a normal app, full screen, with voice and notifications.

If you use ngrok a lot, claim its free static domain so your link never changes. A fixed address is easier to use and easier to keep locked down. See docs/SECURITY.md.

Pick your AI model

Aether works with DeepSeek (the default), OpenAI, Claude, or a model on your own computer. A quick guide:

• Local model: free and fully private. Our top pick if you have a decent GPU.
• DeepSeek: the budget cloud option. Even a heavy user spends about 2 dollars a month, less than the coffee you drink while it works.
• Claude or OpenAI: the sharpest, and the priciest. For when money is not the question.

Switch any time by editing one line. See docs/PROVIDERS.md.

Privacy and security

Aether runs on your own machine. The only thing that leaves it is the AI request to the provider you picked, and you can avoid even that by running a local model. Your login protects everything, the assistant runs in a sandbox, and risky commands always ask before they run. If you plan to open it to the internet, read docs/SECURITY.md first.

What you need

• A Linux desktop. KDE, GNOME, XFCE, and others all work, and the installer adds the tools yours needs. KDE Plasma gets the most polish.
• Docker
• Google Chrome (only for the YouTube features)
• An AI key from DeepSeek, OpenAI, or Claude, or a model running locally (no key needed)

Development

The backend ships with a unit-test suite covering the parts that most need it — the command safety classifier (which decides what shell commands may run on your machine), the agent's JSON parsing, LLM provider resolution, authentication, and the skill-catalog invariants. To run it:

cd backend
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt -r requirements-dev.txt
pytest

The same suite runs on every push and pull request via GitHub Actions (.github/workflows/ci.yml).

---

Made for people who want a computer that listens, on their own terms. If you like it, a star helps others find it.