Clients

Clients are what you talk to Jarvis through: voice nodes, the mobile app and the desktop agent. Every client connects to one server, jarvisd, the Jarvis Automation server you installed on a Mac, a Windows PC or a Linux machine (see Install).

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.

graph LR
    Node["Voice node<br/>(Raspberry Pi)"] --> Server["jarvisd<br/>your server"]
    Docker["Node in Docker"] --> Server
    Mobile["Mobile app"] --> Server
    Agent["Desktop agent"] --> Server
    Mobile -.->|sets up| Node

One server address

A client needs only the server's address: http://<server>:7700, where <server> is the IP address or hostname of the machine running jarvisd. Everything else (sign-in, voice, spoken replies, notifications) is served by the same jarvisd and found from that address.

Port	What it serves	Who uses it
7700	The address clients start from: it tells them where everything else is	Nodes, the mobile app, the desktop agent
7710	The web app, and the setup wizard at http://<server>:7710/setup	You, in a browser

Sign-in and the other services are served on 7700 too, each under its own path (see One API port), and jarvisd runs its own MQTT broker for messages to nodes on 1884; clients find all of them through 7700, so you never type them. The rc18 release still serves sign-in on its own port, 7701; later releases serve it on 7700 under /auth, and clients find it either way. To check that nodes and phones can reach the server, run jarvisd doctor on it; if a firewall is in the way, sudo jarvisd doctor --fix opens the ports.

jarvisd advertises itself on the local network over mDNS as _jarvis-config._tcp. The desktop agent's first-run setup lists the servers it finds that way. The mobile app asks for the address once, at Set Server URL on its first screen, and hands it to every voice node it sets up.

Client types

Voice node on a Raspberry Pi

The main voice interface: a Raspberry Pi with a microphone and speaker that listens for "Hey Jarvis", sends what you say to the server and plays the reply. The Python node (jarvis-node-setup) is the supported one. A Go node (install-pi.sh from each server release) is an experimental preview.

Mobile app

The Jarvis Automation app for iOS and Android. Chat with Jarvis, set up voice nodes over Wi-Fi, manage devices, routines and settings, and get notifications and the inbox.

Provisioning

How a fresh Pi node is set up headless: it opens a Wi-Fi hotspot (jarvis- and 8 characters from its MAC address) and waits for the mobile app to give it your Wi-Fi, its keys and the server's address.

Desktop agent

Jarvis Automation Agent runs on your own Mac, Windows or Linux computer: installers, pairing with a code, the macOS permissions, dictation, and the Jarvis Browser Control extension (which needs its website access turned on).

How clients authenticate

Client	Auth method	Header
Voice node	Node API key	X-API-Key: {node_id}:{api_key}
Mobile app	JWT bearer token	Authorization: Bearer <token>
Desktop agent	Node API key, issued when its pairing code is approved	X-API-Key: {node_id}:{api_key}
Provisioning	Provisioning token (one-time)	Via registration payload

Coming from the old install

Before jarvisd, each part of the server was its own service on its own port (config service, auth, command center, TTS, notifications) and clients reached them one by one. That install is retired for new users; its pages are frozen under Legacy. An upgraded install keeps the old per-service ports until the server.legacy_ports setting is turned off (see One API port), so existing nodes and the app keep working once they point at it; see Migrating from the old install.