Docs

Getting started

What you set up, what you need, and the shortest path.

What you are setting up

Avni has three parts, and each has its own page.

  • The server (avni-core). A small Rust program that listens for your phone, asks a speech-to-text and a language-model provider, keeps the spending cap, and remembers what you tell it to. Run the server.
  • The link. Your phone has to reach the server from anywhere. Tailscale makes that private and simple. Reach it from your phone.
  • The app. The Android app listens, speaks and does the simple commands itself. Install the app, then pair it.

What you need

  • A machine that stays on: a small Linux VPS, a home server, or your own PC while you try it.
  • An API key for a speech-to-text and a chat model that speak the OpenAI-compatible API. I use Groq's free tier. Model names and limits change, so look up the current ones in the provider's docs.
  • An Android 10 or newer phone (the app targets Android 17 and supports back to 10).
  • A free Tailscale account, if you want it to work away from home.

Pick how to run the server

Docker
Anything with Docker: a VPS, a home server, Docker Desktop. One command, upgrades are a rebuild.
systemd
A small Linux VPS. A hardened service and an install script, no container.
From source
Development on your own PC. One script loads your key and prints a QR code.

The shortest path

On your own PC, with a Groq key in .env, this is the whole thing:

terminal
cp dev/avni.example.toml dev/avni.toml   # then put current model ids in it
just up                                  # loads your key, makes a token, prints a QR code, starts core

Install the app, scan the QR code with the phone's camera, and tap the orb. Away from home, put Tailscale on both ends first. Each step is explained on the next pages, with the other routes beside it.