ElevenLabs Realtime Integration

Use ElevenLabs Realtime for low-latency voice conversation with ElevenLabs voices, in one connection.

ElevenLabs Realtime runs speech recognition, the conversation and the ElevenLabs voice behind a single realtime connection. For your assistant it works like the other realtime providers: one model handles the whole voice pipeline, so separate STT and TTS settings do not apply.

There are two models:

Model	Best for
`eleven_realtime_v1_max` (default)	Receptionists that take down details such as names, addresses and phone numbers, long or detailed instructions, and assistants that rely on several tools. It thinks before it answers and tells the caller it is checking something while it does.
`eleven_realtime_v1_mini`	The fastest answers and the lowest cost, for short and simple conversations.

Setup

ElevenLabs Realtime runs on VoiceDock's platform keys by default, so you can select it without adding an API key first.

1. (Optional) Add your own ElevenLabs API key

Only needed to run it on your own ElevenLabs account. It uses the same ElevenLabs key as ElevenLabs text-to-speech and Scribe. Add it under Integrations → API Keys.

2. Configure the assistant

{
  "llm_config": {
    "provider": "elevenlabs_realtime",
    "model": "eleven_realtime_v1_max",
    "voice": "brian",
    "language": "nl",
    "turn_detection": { "eagerness": "medium" },
    "messages": [
      { "role": "system", "content": "You are the receptionist of Acme. Keep your answers short." }
    ]
  }
}

Field	Values	Notes
`model`	`eleven_realtime_v1_max` (default), `eleven_realtime_v1_mini`
`voice`	an ElevenLabs voice id, or a named voice such as `brian` (default)	Any voice from your ElevenLabs voice library.
`language`	a language code such as `nl` or `en`	The language callers speak. Leave it out to let the model detect it.
`turn_detection.eagerness`	`low`, `medium` (default), `high`	How soon the model decides the caller has finished. `high` answers sooner, `low` waits longer for a caller who pauses mid-sentence. In the dashboard this is Turn taking in the language model panel.

Set the language. The model recognises the caller better when it knows which language to expect, and it answers in that language. For an assistant that only takes calls in one language, always set it.

Tools

Your tools in llm_config.tools work on this provider: webhooks, end_call, transfer_call and the tools of an MCP server. For an assistant that depends on tools for most of its answers, such as booking appointments, eleven_realtime_v1_max is the more reliable choice. Test both with your own scenarios in the Arena before you go live.

How it behaves

The model owns the turn-taking. It decides when the caller is done; turn_detection.eagerness sets how soon. speech_config.turn_taking does not apply, as on every realtime provider.
The first message and silence prompts are spoken as given. The greeting, silence prompts and the outbound opening line are passed to the model as a line to say.
There is no text-to-speech step. Pronunciation replacements and a fixed closing line (end_call_message) are pipeline-only and are refused on this provider, as on the other realtime providers.
Summary and analysis after the call run on an OpenAI text model, as ElevenLabs has no text API for that work. Choose another model per assistant with postcall_llm_config.

Pricing

Platform keys (default): ElevenLabs usage at cost + €0,07/min orchestration. ElevenLabs charges per minute plus the tokens of the model, and reports the cost of every conversation; that reported amount is what you are billed. Max uses more tokens than Mini.
Bring Your Own Key (optional): €0,07/min orchestration. ElevenLabs bills your own account.

Data processing

ElevenLabs keeps a record of each conversation (transcript and audio) in the account of the key that was used. On VoiceDock's platform keys we delete that record once the cost of the call has been read. On your own key it stays in your ElevenLabs account, under your own ElevenLabs settings. The sub-processor list on the Trust Hub has the details.

API Reference

See BYOK API Reference for managing your ElevenLabs API key.

ElevenLabs Realtime Integration

On this page