Your own devices as a private AI cluster

The most capable AI computer you own is probably asleep in the next room.

Storyline’s next bet is to use it. Your iPhone, your iPad, and your Mac working as one private cluster over your own local network, with no cloud in the middle.

The idea started with an honest split. AI in writing does two very different jobs. One is help in the moment while you type, which has to be instant and light. The other is background labour: a critique of a whole novel, a continuity audit across twenty chapters, work measured in minutes rather than milliseconds. Most apps pretend these are the same job. They are not.

So map them onto the devices you already have. The phone and iPad are the writing surface, where the assistant is immediate and present. The Mac is the workshop, where the heavy passes run. They are stitched together by your own iCloud account and, when the devices share a Wi-Fi network, a direct encrypted link between them. There is still no Storyline server anywhere in this.

The interesting part is what decides where a task runs. It is not a model. It is a small, deterministic rule engine, the kind of thing you can step through in a debugger. It looks at the task, at what this device can do right now, and at what your other devices can do right now, and it picks one of four answers: run it here, hand it to a more capable device over the local network, queue it in your iCloud for whichever device wakes up first, or refuse and tell you exactly why.

That last answer is the one I care about. When your Mac is asleep and a task needs it, the wrong move is to quietly run a weaker model and hand you a worse result you do not understand. The honest move is to say: this pass needs your Mac, which is offline, want me to queue it for when you open it. The models are the smart part. The placement of the models is deliberately dumb and predictable, because you should always be able to trust where your work went.

The network under it is local-Wi-Fi-first. Your devices find each other over Bonjour, confirm they are yours by checking they share the same iCloud identity, and hold an encrypted channel open between them. On the same network a handoff is tens of milliseconds, fast enough that reaching your Mac beats running a weak model on your phone. Off the network it falls back to your private iCloud. A sleeping Mac is not an error, it is a worker that catches up when you wake it.

The transport and the encryption for this already exist in the codebase, and I just put them through a full security review. The cluster on top is a phased build, so I am being careful to call it a bet, not a feature. But the shape is clear: the Apple hardware you already own becomes your own AI workshop, and nothing about your manuscript has to leave it.

Still early. More soon.