Setup wizard¶
Once the server is running, open http://<host>:7710/setup in a browser (<host> is the machine running the server, or localhost on that machine) to finish setting up. The installers open it for you.
Jarvis Automation is a pre-release (release candidate rc18.17). Builds are on the releases page; each one is marked Pre-release, so open the newest and look under Assets.
The setup link and its token¶
The link the installer opens looks like this:
The part after #token= is the setup token: a one-time secret that proves you can reach the server itself, so nobody else on your network can create the first admin account before you. The server writes it to setup-token in its data directory on first start, readable only by the server's account, and deletes it once the first admin account exists.
To print the link again, run this on the server:
--qr draws the link as a QR code in the terminal: scan it with your phone to open the wizard there, which suits a server without a screen. Once setup is done, jarvisd setup-link prints the admin's address instead. On Windows, run it in an administrator PowerShell without sudo; on Linux with a --user install, without sudo.
From another device, use the server's address, not localhost: for example http://192.168.1.20:7710/setup#token=….
The steps¶
The wizard saves each step as you go: close the tab and the link opens where you left off.
- Check: checks that phones, voice nodes and other computers can reach the server. If something is blocked (usually the firewall), it shows the check that failed and its fix.
- Account: creates the first account, the household's admin.
- Hardware: shows what the server found on this machine, which decides the models it can run.
- Language model: understands requests and writes the answers; the biggest download. The wizard suggests one that fits and marks the others Fits or not; see Requirements and choosing a model.
- Speech-to-text: turns what you say to a voice node into text. Needed for voice.
- Voice: how Jarvis sounds when it answers. Needed for voice.
- Voice recognition (optional): tells household members apart by voice.
- Memory & search (optional): lets Jarvis find related memories and notes. Small, and runs on the CPU.
- Privacy: asks what may leave the server, such as add-on commands from the Pantry and push notifications to phones.
- This computer (optional): Also use Jarvis Automation on this computer downloads the desktop agent for the computer you are on, pairs it with the server and, if you tick it, sets up the browser extension. Skip for now leaves it for later; the admin's Add-ons page has the same download.
- Done: the wizard hands over to the admin.
Steps 4 to 8 each pick the model for one job and suggest one that fits this machine. Nothing large is downloaded until you confirm a step; the download then carries on while you move to the next one. Install everything recommended, on the first of them, confirms all five suggestions at once. Each can be skipped (Skip for now on the three Jarvis needs, Skip on the two optional ones) and set later on the admin's Models page.
After setup, the admin stays on port 7710 (http://<host>:7710). Phones and voice nodes need the server's address, http://<host>:7700: you type it into the mobile app, and the app passes it to a voice node when it sets the node up. Only the desktop agent finds the server on the network by itself.
Next¶
- Add a computer: the desktop agent.
- Add a voice node and the mobile app.
If the wizard doesn't open, or opens on the server but not from another device, see Troubleshooting.