The setup, the one prompt that turns your messages into a reply list, a to-do list and a calendar check, and the six searches worth saving.
OpenAI added an Apple Messages plugin, and once it is on, ChatGPT can search every text you have ever sent or received, catch you up on a conversation, draft replies, and send them once you approve.
Most of the coverage is about having it write your texts, which is honestly the least interesting thing it does. The real unlock is that years of messages become searchable by meaning instead of by scrolling.
Check this first, or you will waste ten minutes. It runs in the ChatGPT desktop app on Apple Silicon Macs, meaning M1 and newer. It does not work on Intel Macs, on the web, on your phone, or in the Codex CLI. It is also off until you switch it on.
1. Open the ChatGPT desktop app on your Mac and go into Work or Codex.
2. Open Plugins, look under Public, find Apple Messages and install it.
3. Approve the Mac permissions it asks for. It needs Full Disk Access so it can read the messages database, access to your contacts so names mean something instead of phone numbers, and permission to control the Messages app so it can actually send.
4. Type @Messages at the start of any prompt to pull it in.
What it can actually see. It covers iMessage, SMS and RCS, so green bubbles work too. It reads your conversations and can send through the Messages app. It does not read anything from WhatsApp, Instagram or Signal.
This is the one worth saving. It reads a week of messages and hands back three lists: everyone waiting on you with a draft reply already written, everything you promised somebody and forgot, and every date anyone mentioned that is still coming up.
The second list is the one that gets people. Nobody remembers the "yeah I'll send that over" they typed nine days ago.
These are the ones I would actually use. Copy the whole block and keep it somewhere, then swap the names.
By default ChatGPT shows you the message and the recipient and waits for you to approve before anything sends. At some point it will offer to always allow sending to a particular chat, which turns that approval off for that person.
Say no. Your inbox is text other people wrote, which means anyone who messages you could put instructions inside a message aimed at the assistant. OpenAI's own documentation says to keep per-send approval on for conversations that might contain untrusted or misleading instructions. Approving each message takes a second and it is the only thing standing between a weird text and something going out under your name.
One more practical note. If you have a task set to full access, or you have otherwise turned approval prompts off, the plugin can end up unable to show you the confirmation it needs, so the message just never sends. If a text seems to vanish, that is usually why.
If you are setting ChatGPT up properly, this walks through the rest of it.