Voice
ExoHuman offers three voice features in the Workspace:
| Feature | What it does | Powered by |
|---|---|---|
| Voice call | A real-time spoken conversation with an agent | Gemini Live |
| Speak replies aloud | Reads each written reply in the agent’s voice | ElevenLabs |
| Dictation | Turns speech into text in the message field | ElevenLabs, or your browser |
Phone calls are covered in Voice & telephony.
Voice calls
Section titled “Voice calls”- Open the Workspace, pick an agent and the chat the call should belong to.
- Click the call button, the leftmost control in the composer.
- Talk. The orb shows the call state (Listening, Speaking, Working) and the agent’s canvas appears beside it.
- End with End call or Esc. M mutes.
When you ask for real work during a call, the agent hands it off as a task that runs as the agent, in that chat, with its skills, files and memory — while the conversation continues. Up to three tasks can be in flight. When one finishes, the agent brings up the result at the next natural pause.
The spoken turns land in the chat as a Voice call block, and the agent’s next typed turn knows what was said. Anything the agent showed stays on its canvas.
Limits
Section titled “Limits”- A call lasts up to 30 minutes; the agent is told to wrap up shortly before the limit.
- Calls are available to signed-in platform users, not to external email-code clients.
- A call needs microphone permission on your browser.
Requirements (admins)
Section titled “Requirements (admins)”- A Gemini API key in Settings → Integrations. See Platform keys.
- Each agent has a voice (default Kore) used for calls and phone calls, set on the agent.
- An agent can have a dedicated voice persona by placing
voice-agent-system-prompt.mdin its workspace. - The agent’s
template.yamlcan narrow the tools available during a call.
Speak replies aloud
Section titled “Speak replies aloud”When an agent has an ElevenLabs voice configured, a speaker button above the composer reads each reply aloud on your device. Admins add the ElevenLabs key in Settings → Integrations → Voice replies.
Dictation
Section titled “Dictation”The microphone in the composer types what you say into the message field — nothing is sent until you press Enter. It uses ElevenLabs speech-to-text when the instance key allows it, and your browser’s speech engine otherwise. If neither is available the microphone is hidden. Dictation works for external clients too.