Is there an iMessage bot API?

If you have built a bot for Telegram, Slack or Discord, the pattern is familiar: register an app, get a token, subscribe to a message event, reply. Every platform ships this. iMessage does not. There is no bot registration, no token, no event subscription, and no Apple-run webhook.

That gap has become much more visible in the last two years, because the obvious place to put a conversational AI agent is the app people already have open. So the question gets asked constantly, and the answer has not changed.

What a bot needs, and where it has to come from

Strip a messaging bot down and it needs exactly four things. On Telegram, all four come from Telegram. On iMessage, all four have to come from somewhere else.

The shape of an iMessage bot that works

Once you have a provider, the loop is conventional. This is the whole thing — an inbound webhook, your logic, an outbound send.

// Illustrative. Field names differ per vendor — check their live docs.
import express from "express";
 
const app = express();
app.use(express.json());
 
app.post("/webhook/imessage", async (req, res) => {
  // 1. Acknowledge fast. Providers retry on a slow or failed response,
  //    and a duplicate reply is worse than a late one.
  res.sendStatus(200);
 
  const { from, text, message_id } = req.body;
 
  // 2. Deduplicate. At-least-once delivery means you WILL see repeats.
  if (await seen(message_id)) return;
  await remember(message_id);
 
  // 3. Your actual logic.
  const reply = await answer(text);
 
  // 4. Send. Every vendor has some version of this call.
  await fetch("https://api.example-provider.com/v1/messages", {
    method: "POST",
    headers: {
      Authorization: `Bearer ${process.env.IMESSAGE_API_KEY}`,
      "Content-Type": "application/json",
    },
    body: JSON.stringify({ to: from, text: reply }),
  });
});
 
app.listen(3000);

Note what is doing the work here: acknowledging before processing, and deduplicating on the provider's message id. Both are consequences of at-least-once delivery, which every provider in this category uses. Skip them and your bot will double-reply in production even though it never did in testing.

If the bot is an AI agent

Most people asking this question in 2026 are putting an LLM behind the webhook. That works, and several providers now position themselves specifically at agent builders — Photon and Linq both lead with it. Two things change when the responder is a model rather than a state machine.

• Latency becomes a product decision. A model that takes eight seconds to answer reads as a person who wandered off. Send a typing indicator if your provider supports one, and set a hard timeout with a graceful fallback message.
• The model will eventually say something you would not. A blue bubble carries the implication of a person at your business. Put a real content filter and an escalation path in front of it, not just a system prompt. AI agents and customer texts covers this properly.

For the language-specific starting points, see Node.js and Python.