Michael Di Giacomi
All posts

Originally shared on LinkedIn

Building an iMessage bridge for my OpenAI dot

My dot could send an iMessage from a Mac. The missing piece was knowing when someone replied.

I wanted my OpenAI dot to be able to follow a conversation in iMessage. It could use Messages on my Mac to send a text, which was a good start. But in my setup, when someone replied, the dot didn’t get notified. I still had to tell it to go check.

OpenAI has a built-in texting feature, too. It lets you text your own dot over iMessage or RCS. I wanted something different: for a few approved people to be able to reply in iMessage and have my dot know to check for them.

That’s the problem we built the bridge to solve. The Mac looks for new messages in a small set of approved one-to-one chats. When a message arrives, the bridge sends the dot a heads-up. The alert has a conversation alias, an event ID and a timestamp, but no message text or attachments. Reading the actual message and replying are separate, authorized steps on the Mac.

I set up a separate Apple account for the assistant, so this doesn’t connect to my own inbox. The people I’ve approved can use the app they already have, without the bridge getting access to every conversation.

We’ve tested the whole trip: I sent a message, the dot got the notification, and I received its reply. The harder part has been making that dependable. We’ve seen delays and restart issues. And if the Mac loses its connection halfway through a send, the bridge needs to remember that it might already have sent the message. Retrying blindly is a good way to text someone twice.

It’s still a work in progress. But the original idea was pretty simple: I shouldn’t have to tell an assistant that someone replied before it can follow the conversation.

OpenAI’s guide to texting your dot

Read the technical follow-up: how the bridge works