n8n Voice Agents: Connect AI Phone Systems to Your Workflows

Voice Agent Architecture

A voice agent system has three layers:

Voice layer: Handles the phone call. Converts speech to text, manages the conversation flow, and converts text back to speech. This is where Vapi or Retell AI operate.

Intelligence layer: The AI brain. Decides what to say based on what the caller said. Claude API or other LLMs handle this.

Integration layer: Connects calls to your business systems. When a lead qualifies, create a CRM record. When an appointment is booked, update the calendar. n8n handles this.

Vapi + n8n Integration

Vapi provides the voice layer and handles much of the complexity:

• Phone number management
• Speech-to-text (STT)
• Text-to-speech (TTS) with ElevenLabs or other voice providers
• Real-time conversation management
• Webhook events for integration

Setting up Vapi webhooks to n8n:

In Vapi, configure webhook URL to your n8n webhook endpoint.

Vapi sends events:

• call.started — new inbound call
• call.ended — call completed (with transcript)
• tool.call — your assistant is calling a tool you defined
• status-update — conversation progress

n8n handles the events:

1. Webhook trigger: Vapi event received
2. IF node: which event type?
3. Branch for call.ended: process transcript, extract data, update CRM
4. Branch for tool.call: execute the tool request and return result

Building an Inbound Lead Qualification Agent

The agent's goal: Answer inbound calls, qualify the caller as a potential customer, and book a meeting if qualified.

Vapi configuration:

n8n workflow for tool calls:

1. Vapi webhook: tool.call event with function.name = "book_meeting"
2. Extract: name, email, slot from tool arguments
3. Calendly API: create booking
4. HubSpot: create contact + deal
5. Email: send confirmation to caller
6. Return: confirmation message to Vapi → agent reads to caller

Outbound Calling Campaign

Use n8n to trigger outbound calls:

1. Airtable trigger: new row (leads to call)
2. HTTP: Vapi API → initiate outbound call

1. Wait: 5 minutes (don't flood calls)
2. Continue loop for next lead

Post-call processing:

1. Vapi webhook: call.ended (with transcript)
2. Claude: analyse call outcome — interested, not interested, wrong number, voicemail, callback requested
3. HubSpot: update contact with call outcome
4. If interested: create follow-up task for human rep
5. If voicemail: schedule retry in 24h

Appointment Reminder Calls

Simple voice reminder workflow:

1. Schedule trigger: check appointments for tomorrow
2. Database: get appointments with phone numbers
3. HTTP: Vapi API → initiate reminder call for each
4. Vapi handles: "Hi, this is a reminder about your appointment tomorrow at 2pm with [Company]. Press 1 to confirm, 2 to reschedule."
5. Vapi webhook: call.ended → n8n processes response
6. Calendar: update attendance status

Call Analysis and Intelligence

Every call produces a transcript. Extract value from it:

1. Vapi webhook: call.ended with full transcript
2. Claude: "Analyse this sales call transcript. Extract: customer pain points mentioned, objections raised, sentiment, likelihood to convert (1-10), and recommended next step."
3. HubSpot: update deal with AI analysis
4. If likelihood ≥ 7: Slack alert to sales rep
5. Weekly batch: aggregate call insights → email report to sales leadership

Recommended Tools

• n8n — Voice agent workflow orchestration
• Vapi — Voice AI platform; best for building phone agents
• Retell AI — Alternative voice AI platform
• ElevenLabs — Natural-sounding AI voices
• Claude API — AI brain for voice agents
• Twilio — Phone number and telephony infrastructure