Skip to content
Docs0.9GitHubInstall

Using itVoice mode and Listen

Voice mode and Listen

Have a reply read aloud with Listen, or hold a hands-free conversation with an agent in voice mode, in the web client and the iPhone app.

Two ways to hear your agents in the apps, for anyone using the web client or the iPhone app. Listen reads one reply aloud. Voice mode is a spoken conversation: you talk, it sends when you pause, reads the reply, and listens again.

For voice notesVoice note. A recording you send as a prompt. It reaches the agent as its transcript., transcription setup, and dictationDictation. Speaking at your desk into the focused agent or window, through a desktop key. at the desk, see Voice.

Where
Web client Hover a reply and press the speaker button, Listen, or pick Listen from the reply’s menu
iPhone Touch and hold a reply, then Listen

Reading starts after the first piece is ready, so a long reply does not wait for the whole of it. Listen is there in Simple and Advanced alike.

Listen reads the prose of a reply and leaves out what does not work spoken:

  • code blocks, tables, quotes, and tool-call lines are skipped;
  • thinking is never read;
  • a link is read as its text, and a bare address as the word “link”.

A reply with nothing left to read says so on iPhone: “Nothing to read aloud”.

A bar titled Reading aloud shows in the reply’s chat, with a line of the text being read.

Control Does
Play / Pause Pauses and goes on
Back 15 seconds, Forward 15 seconds Skip within the reading
The scrubber Shows elapsed and remaining time; drag to move
Speed Cycles 1×, 1.25×, 1.5×, 2×. The device remembers it.
Stop Ends the reading and closes the bar

The scrubber and the 15-second skips need the server’s voice. With the device’s own voice, the back button becomes From the start and forward is off.

Listen reads one reply and stops. It keeps playing if you open another chat; the bar stays in the chat it belongs to. One sound plays at a time: starting another reading, a video, or voice mode stops it.

There is no voice picker in the apps. The adminAdmin. The person who installed Vendetta and runs it. People are added under the admin's account. chooses:

  • With [plugins.speech.narration] set on the computer, replies are read in that voice: a local program such as Piper, or an ElevenLabs voice. Each device caches what it has played, so a replay is instant.
  • Without it, the device reads with its own voice. The iPhone picks its best installed voice for the reply’s language; the browser picks a voice for its language, or a Cyrillic one for mostly Cyrillic text.

If the server’s voice fails or a reply is over its limit, the device’s voice reads instead, with no message.

The admin can have a small model rewrite each reply for the ear before it is read. Nothing in the apps turns it on or off. When the rewrite fails before anything was heard, the reply’s own prose is read. Reading aloud has the setting, [plugins.messaging.narration.rewrite].

Voice mode needs transcription set up on the computer; without it, its buttons are not shown. It is an Advanced feature: Simple keeps voice notes and hides voice mode.

Where
Web client Voice mode in the chat’s header or its menu, the Voice chip above the message box, or Voice mode in a chat’s menu in the list. While a turn runs, the button is on the working line.
iPhone Voice Mode in the chat’s menu, or Voice by the message box (on the working line while a turn runs)

A panel replaces the message box: an orb that moves with your voice, a status line, and a hint.

The panel says What is happening
Speak now It is listening and has not heard you yet
Listening It hears you. Pause when you are done; it sends by itself.
Sending What you said goes to the agent as a voice note
The agent is working It waits for the reply
The agent is speaking It reads the reply, then listens again
Paused The microphone is not allowed. Allow it and start again.
The agent needs you on screen See below
  • It sends after about 1.4 seconds of quiet, judged against the room’s background noise. Send now sends at once.
  • One turn of speech is sent after 90 seconds even if you have not paused.
  • What you say arrives as a voice note with its transcript, like any other.
  • Every final reply the agent gives in that chat is read. If the agent is still working after a reply, voice mode waits for the next one; once the agent stops, it listens again.

Voice mode stays with its chat while you look elsewhere. In another chat a small pill shows the chat’s name and what voice mode is doing; on other screens the panel floats. On iPhone, tap the name to go back to the chat.

  • Press the close button, End voice mode.
  • Drag the panel down.
  • In the web client, press Esc.

Voice mode does not answer questions or approvals for you.

The agent Voice mode says Then
Asks a question “The agent has a question for you on screen.” Shows The agent needs you on screen
Needs an approval “The agent needs your approval in its terminal.” Shows The agent needs you on screen

Answer the card in the chat (or the approval where it is asked). Voice mode stays on that status and does not listen or read further; end it and start it again to carry on talking.

If the agent stops or exits without a reply, voice mode goes back to listening.

Talk to Master starts voice mode with the Master’s chat, without leaving the screen you are on.

Where
Web client Talk to Master at the top of the sidebar; in jump mode, V
iPhone The waveform button on the home screen

It needs transcription, as voice mode does, and is hidden in Simple. See Master, managers, teams for what the Master can do.

Limit
One reading 20,000 characters of read text; longer replies fall back to the device’s voice
One piece of a reading 4,500 characters
Speeds 1×, 1.25×, 1.5×, 2×
One spoken turn in voice mode 90 seconds
Quiet before a turn is sent About 1.4 seconds
A reading for a person through the admin’s voice 40,000 characters, two readings open at once