Apple introduces Core AI for custom on-device models
What was announced?
According to reporting by MacRumors, Apple introduced Core AI at WWDC26: a new framework for running custom on-device models on Apple silicon. Two details stand out in the coverage. Models are compiled ahead of time. And Apple ships Python tooling for converting PyTorch models, which is where most open-weight and research models live today.
Foundation Models is Apple handing developers its model. Core AI is Apple letting them bring their own — and treating that path as supported rather than tolerated.
What does it mean for personal, on-device writing?
Custom local models have been an afterthought on iOS. The serious on-device model work happened in hand-rolled inference stacks that fought the platform. If Core AI is what the coverage describes, that changes: a custom model becomes a supported citizen on Apple silicon, with a sanctioned route from PyTorch to the phone.
For personal writing, the interesting word is custom. A general model with your preferences layered on top is one architecture. A small model actually shaped by you — adapted from your own writing, living on your own device — is another. The second has mostly stayed a research demo because deployment was hostile. Core AI lowers that ceiling. It does not put a personal model on your phone tomorrow. It makes the path legible.
Where does ILURA stand?
The honest version: ILURA does not train model weights, and Core AI does not change that this year.
ILURA personalizes the other way around. Your corrections become rules and tone in a role playbook — your Manager voice, your Founder voice — and the on-device Apple Intelligence model writes under those rules. Training by correction, not by gradient. The result is inspectable, editable, and runs on today’s hardware.
Core AI matters to us as a ceiling-lowerer. An eventual on-device personalization step — something small, learned from your edits, living next to your playbooks, never leaving the phone — gets more plausible when the platform supports custom local models properly. That is a research direction, not a near-term feature, and we are saying exactly that. What does not move: no account, Data Not Collected, and writing that happens on your device.
Read the signal through ILURA
Platform news matters when it changes what users expect from personal AI. ILURA reads these shifts through one lens: private agents trained by the user on iPhone.
- What becomes possible?
- What should stay user-controlled?
- What belongs on device?
Try it now
Put this to work on a real message.
Open ILURA, bring in a message you actually need to handle today, and get it done in your voice — free, on device, no account. It learns the preference, so the behavior carries to the next one.
Free to start · No account · Data Not CollectedQuick answers
- What is Apple's Core AI framework?
- Per MacRumors' coverage of WWDC26, Core AI is a new framework for running custom on-device models on Apple silicon, with ahead-of-time compilation and Python tooling that converts PyTorch models for local execution.
- Will ILURA train a custom model on my phone?
- Not in the near term. ILURA personalizes through playbooks — rules and tone learned from your corrections — not by training model weights. Core AI makes an eventual on-device personalization step more plausible, but today it is a research direction.