Skip to main content
aifeatures.ai
Ops teams turning recordings into actions

The async voice pipe

Whisper transcribes, Groq decides the tool calls, ElevenLabs speaks the reply, n8n writes calendar and CRM. Not a live PSTN loop.

Who owns what

  1. Whisper

    The transcript — inbound audio to text. Not intent and not the phone trunk.

    Try Whisper (opens in a new tab)

  2. Groq

    The fast decide — structured tool calls from that text. Not telephony and not SMTP.

    Try Groq (opens in a new tab)

  3. ElevenLabs

    The one-way reply — WAV from the decision. Not the live interrupt loop.

    Try ElevenLabs (opens in a new tab)

  4. n8n

    The side effects — calendar, mail, CRM after the model returns. Not raw audio.

    Try n8n (opens in a new tab)

How to use this stack

  1. Webhook or voicemail dump lands a recording. Carrier and SIP stay infra.
  2. Whisper transcribes. Groq returns intent plus allowed tool names — book, FAQ, escalate.
  3. n8n executes the write. ElevenLabs renders a confirmation the caller can hear later.
  4. Do not sell this as ten concurrent live calls. That needs a realtime layer (Vapi / LiveKit), which we do not track.

Why we picked this: The inbound voice call agent: ChatGPT Realtime, not a chatbot