Build an iMessage Chatbot with Blue Replies and OpenAI

AI-Powered Customer Support via iMessage

Imagine your customers texting your business via iMessage and getting instant, intelligent responses at any hour — no hold times, no support tickets, no waiting until Monday. By combining Blue Replies's iMessage API with OpenAI's language models, you can build a chatbot that handles common inquiries, answers FAQs, and seamlessly escalates to human agents when needed.

This is not a clunky phone tree or a "type 1 for billing" bot. This is natural, conversational AI delivered through the channel your customers already prefer.

Architecture Overview

The system works in three steps:

• 1. Customer sends iMessage → Blue Replies receives it and triggers your webhook
• 2. Your server processes the message → Sends the message to OpenAI with context and instructions
• 3. OpenAI generates a response → Your server sends it back via Blue Replies as an iMessage

Step 1: Set Up the Webhook Handler

Create an Express.js server that receives incoming iMessages and routes them to OpenAI:

import express from 'express';
import OpenAI from 'openai';
import Blue Replies from '@bluereplies/sdk';

const app = express();
app.use(express.json());

const openai = new OpenAI({ apiKey: process.env.OPENAI_KEY });
const bluereplies = new Blue Replies({ apiKey: process.env.BLUEREPLIES_KEY });

// In-memory conversation history (use Redis in production)
const conversations = new Map();

app.post('/webhooks/bluereplies', async (req, res) => {
  res.status(200).send('OK');

  const { type, data } = req.body;
  if (type !== 'message.received') return;

  const { from, body } = data;
  await handleIncomingMessage(from, body);
});

app.listen(3000);

Step 2: Build the AI Response Logic

Maintain conversation history for context-aware responses:

async function handleIncomingMessage(from, body) {
  // Get or create conversation history
  if (!conversations.has(from)) {
    conversations.set(from, []);
  }
  const history = conversations.get(from);

  // Add the customer's message to history
  history.push({ role: 'user', content: body });

  // Send typing indicator while AI processes
  await bluereplies.messages.sendTypingIndicator({ to: from });

  // Generate AI response
  const completion = await openai.chat.completions.create({
    model: 'gpt-4o',
    messages: [
      {
        role: 'system',
        content: `You are a helpful customer support agent for Blue Replies,
          an iMessage API platform. Be friendly, concise, and helpful.
          Keep responses under 300 characters when possible — this is
          a text message conversation, not an email.
          If you cannot answer a question, say you will connect them
          with a human agent.`,
      },
      ...history,
    ],
    max_tokens: 200,
  });

  const reply = completion.choices[0].message.content;

  // Save AI response to history
  history.push({ role: 'assistant', content: reply });

  // Send the response via iMessage
  await bluereplies.messages.send({ to: from, body: reply });

  // Check if escalation is needed
  if (reply.includes('connect you with') || reply.includes('human agent')) {
    await escalateToHuman(from, history);
  }
}

Step 3: Add Smart Escalation

Not every question should be handled by AI. Build escalation logic for complex issues:

async function escalateToHuman(customerPhone, history) {
  // Notify your support team via Slack, email, or CRM
  const summary = history
    .map(m => `${m.role}: ${m.content}`)
    .join('\n');

  await fetch(process.env.SLACK_WEBHOOK_URL, {
    method: 'POST',
    headers: { 'Content-Type': 'application/json' },
    body: JSON.stringify({
      text: `🔔 Escalation needed for ${customerPhone}\n${summary}`,
    }),
  });

  // Send handoff message to customer
  await bluereplies.messages.send({
    to: customerPhone,
    body: 'I am connecting you with a team member who can help further. They will reach out shortly!',
  });
}

Best Practices for iMessage Chatbots

• Keep responses short: This is texting, not email. Aim for 1-3 sentences per response.
• Use typing indicators: Show the typing bubble while the AI processes to mimic natural conversation flow.
• Maintain context: Use conversation history so the AI remembers what was discussed. Store in Redis with a 24-hour TTL.
• Set expectations: When a customer first messages, let them know they are chatting with an AI assistant.
• Escalate gracefully: When the AI is unsure, hand off to a human rather than guessing. Customers appreciate honesty.
• Add personality: Emojis and casual language feel natural in iMessage. "Let me check that for you! 🔍" feels right.

Going Further

Once your basic chatbot is running, consider adding:

• RAG (Retrieval-Augmented Generation): Connect your knowledge base so the AI can answer product-specific questions
• Order lookups: Let customers check order status by saying "Where is my order?"
• Appointment scheduling: Integrate with your calendar to let the AI book meetings
• Sentiment analysis: Detect frustrated customers and escalate proactively