JARVIS UltraManualv1.31

How-to guides

Talk to your agents

Agents can read their replies aloud, each in its own voice, and in talk mode you speak instead of typing. Voice settings are in Settings > Voice.

Choose a voice engine

Open Settings > Voice and pick an Engine:

Engine What it is
Browser Your device's own voices. Nothing to install; quality depends on the device.
Piper Eight natural local voices that run on your computer, free and offline. Needs a one-time install (below).
ElevenLabs High-quality voices from ElevenLabs. Needs an ElevenLabs key: paste it under ElevenLabs and press Save key.

Then, under A voice for each agent, pick a voice per agent and press Hear it to listen. Speed runs from 0.6 to 1.6. Changes save as you make them. A reply is read up to its first 2,500 characters, without the Markdown.

Install the Piper voices

Piper is installed from a terminal, in the JARVIS Ultra folder:

bash scripts/install-voice.sh

It installs Piper into voice/.venv, downloads eight voices (about 500 MB) into voice/models, and starts a voice service on 127.0.0.1:8769. It ends with Done. Open Settings > Voice: Piper shows eight voices, one per agent. The Local voices (Piper) card then shows the voices as installed and running, and the engine switches to Piper with a voice for each agent. With more than eight agents, some share a voice.

The installer can do the same: ./install.sh --voice (or .\install.ps1 -Voice). On Windows, Piper is untested. If the voice step fails, see Install and service problems.

Read a reply aloud

Press Read aloud under any reply in Chat; Stop reading ends it. Starting one voice stops any other.

Use talk mode

  1. In a conversation, press the microphone button beside Send ("Talk: speak instead of typing").
  2. Allow the microphone when the browser asks. The bar above the message box says "Listening… what you say goes into the message box and is sent when you pause."
  3. Speak. When you pause for about a second and a half, your words are sent. Typing cancels the send.
  4. The reply is read aloud, and listening pauses while the agent speaks. Press Escape to interrupt it.
  5. Press End talk (or the microphone button again) to stop. Opening another conversation ends talk mode too.

With Greet me when talk mode starts on, the agent says hello in its own voice first.

Talk mode uses your browser's speech recognition: Chrome sends the sound to Google to transcribe it, Safari uses Apple. It needs Chrome or Safari, and either the dashboard on this computer or its https Tailscale address; a plain http:// address from another device cannot listen. The Listening card says Works here or why not.