Skip to main content

Voice

You can talk to your agents in the dashboard. Dictate instead of typing, hear replies read aloud, or hold a live voice conversation where you and the agent speak as naturally as on a phone call. A wake word can start one hands-free.

Live voice conversations​

Tap the phone toggle next to the microphone (or slide the mic button left) and just talk. The conversation is full-duplex:

  • Open mic, instant interruptions. The agent hears you the whole time. Talk over a reply and it stops mid-sentence and listens, even while it's silently working on a tool call. Brief acknowledgements like "ok" don't interrupt; real sentences do.
  • A live transcript builds in the chat input as you speak, exactly like dictation, and sends itself when you finish your thought.
  • Spoken replies stay short, while documents, artifacts, and images still render on screen. The agent says a quick line before it reaches for a tool and keeps you posted during long tool runs, so the line never goes dead.
  • The glow around the chat input follows the conversation — rippling with your voice while listening, a violet shimmer while thinking, and the agent's speech waveform while it talks.
  • Click to edit. Tapping into the chat input mutes the mic and holds your transcript as an editable draft. Fix a word and Send dispatches it as a spoken turn, or tap the mic to hand the draft back to the live conversation. The mic button is a real mute toggle throughout: while muted, no audio leaves your device.
  • Agents can hang up. Say "thanks, good night" and the agent closes the session after its farewell. If you ask for something long, it can say it will keep working in the chat and notify you when done, then leave the spoken mode while the work continues.

Voice conversations work on regular chats and on interactive terminal chats, and they survive the agent's process being restarted: the next thing you say picks the conversation back up.

Live conversations need the phone service running on your install — see Setup below.

Wake word: "Hey OtoDock"​

With the wake word enabled, saying "Hey OtoDock" on any dashboard page opens your favorite agent in a fresh voice conversation. Say "Hey" plus an agent's name to get that agent directly; renamed agents are matched automatically, with nothing to train.

Detection runs entirely on your device, as a small keyword-spotting model in the browser. Audio never leaves your machine until a conversation actually starts. It's a per-user opt-in, off by default: turn it on under User Settings → General → Wake word. It listens wherever the dashboard is open in the foreground, and a wake that hears nothing for ten seconds hangs up on its own. Only agents added to your account can be woken.

The wake phrase itself is English-phonetic; the conversation after it speaks your language as usual.

Dictation​

Prefer to type by voice? Tap the microphone and speak. Your words transcribe live into the input, building on anything you've already typed, and you send when ready. Dictation works without the phone service: it uses your device's built-in speech recognition or the platform's speech-to-text provider, depending on the admin policy.

To pick your dictation language, press and hold the microphone button (or right-click it on a desktop) and choose from the menu. Your choice applies from the next time you dictate or start a conversation, and you can also change it under User Settings → Audio.

Read aloud​

Any reply can be played aloud from its speaker icon, in the configured text-to-speech voice — handy when you're away from the screen but not in a full conversation.

Setup​

Voice is configured by an admin under Setup → Audio:

  • Speech providers. OtoDock supports Deepgram and ElevenLabs for speech-to-text, and Cartesia and ElevenLabs for text-to-speech. Create an account and paste its API key into the provider card. Admins pick a voice per language, so each person hears a natural voice in their own language. Supported languages include English, Greek, German, Spanish, French, and Italian.
  • Chat audio policy. Native only (the device's built-in voice, works with no provider connected), Native preferred, or User choice (each person picks under User Settings → Audio, and the choice follows them across devices). An admin can also turn chat audio off entirely.
  • The phone service. Live voice conversations and the wake word run on the same voice engine that handles phone calls: the phone service, included in the standard install (see Installation). Dictation and read-aloud work without it. Dashboard voice needs no phone number or PBX — just the service itself, plus speech providers.
note

Voice usage (transcription time and characters spoken) is tracked alongside your other usage — see Usage & Limits.

Next steps​