How to Build an AI Receptionist with BubblyPhone Agents

An AI receptionist answers incoming calls, understands what the caller needs, and takes action — booking appointments, routing calls, or answering FAQs. With BubblyPhone Agents, you can build one in under an hour using webhooks or real-time streaming.

Prerequisites

A BubblyPhone Agents account — sign up free
A phone number purchased through the platform ($3/month)
A server endpoint that can receive webhooks (or use streaming mode)
An AI/LLM of your choice (OpenAI, Anthropic, Groq, etc.)

Webhook vs Streaming Mode

BubblyPhone Agents supports two modes for handling calls. Choose based on your use case:

Feature	Webhook Mode	Streaming Mode
How it works	Events sent to your endpoint, you respond with actions	Real-time audio bridge connects caller to your AI model
Latency	Depends on your server + LLM response time	Sub-second — audio streams directly
Complexity	Simple — just handle HTTP requests	Simpler — configure model + prompt, we handle the rest
Best for	Custom logic, multi-step flows, tool use	Conversational agents, quick setup

Step 1: Get a Phone Number

Purchase a phone number through the API. This is the number callers will dial to reach your AI receptionist.

curl -X POST https://agents.bubblyphone.com/api/v1/phone-numbers \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "phone_number": "+14155550100",
    "label": "AI Receptionist",
    "webhook_url": "https://your-server.com/webhook",
    "mode": "webhook"
  }'

Step 2: Handle Incoming Calls

When someone calls your number, BubblyPhone sends a webhook to your endpoint. Respond with an action to answer the call and greet the caller.

// Incoming webhook payload
{
  "event": "call.incoming",
  "call_id": "call_abc123",
  "from": "+14155551234",
  "to": "+14155550100",
  "direction": "inbound",
  "timestamp": "2026-04-02T10:30:00Z"
}

Your server responds with an action:

// Your response
{
  "action": "answer",
  "text": "Hello! Thank you for calling. How can I help you today?"
}

Step 3: Process Speech and Respond

After the greeting, BubblyPhone transcribes the caller's speech and sends it to your webhook. Feed it to your LLM and respond.

Tip:For the fastest response times, use a fast model like Groq's Llama or GPT-4o-mini. The caller is waiting on the line — every millisecond counts.

Step 4: Add Tool Use (Optional)

Make your receptionist actionable. In streaming mode, define tools that your AI can call — like booking an appointment or looking up a customer.

// Configure tools on your phone number
{
  "mode": "streaming",
  "model": "gpt-4o",
  "system_prompt": "You are a friendly receptionist for Acme Corp...",
  "tools": [
    {
      "type": "function",
      "function": {
        "name": "book_appointment",
        "description": "Book an appointment for the caller",
        "parameters": {
          "type": "object",
          "properties": {
            "date": { "type": "string" },
            "time": { "type": "string" },
            "name": { "type": "string" }
          }
        }
      }
    }
  ],
  "tool_webhook_url": "https://your-server.com/tools"
}

Summary

You now have a working AI receptionist that can answer calls, understand natural language, and take actions. From here you can:

Add more tools for appointment booking, order lookup, etc.
Switch to streaming mode for lower latency conversations
Set up call recording and transcription for quality assurance
Configure max call duration and silence timeouts