Apple opens up the Foundation Models framework
What was announced?
At the WWDC26 Platforms State of the Union, Apple expanded the Foundation Models framework, the Swift API that lets apps call Apple Intelligence models directly. Four changes matter. The framework now accepts image input. Developers get free access to a model hosted on Private Cloud Compute, with a 32K context window and reasoning. Dynamic Profiles arrive alongside. And a new LanguageModel protocol lets an app put third-party models, such as Claude or Gemini, behind the same Swift API it already uses for Apple’s own.
The shape of the change is worth pausing on. The on-device model, the private-cloud model, and outside models now sit behind one developer surface. An app writes against the protocol; the model underneath can improve or be swapped without the app changing.
What does it mean for personal, on-device writing?
A personal writing tool wants three things from its platform: generation close to the user, privacy it does not have to build itself, and room to grow. This announcement touches all three. The on-device plus Private Cloud Compute stack is now a first-class developer surface. PCC inherits Apple’s platform privacy architecture, so an app using it is not bolting privacy on after the fact. Free access removes the cost argument for defaulting to third-party servers. And a 32K context with reasoning is a different class of capability from the local model alone — enough for long documents, not just short replies.
The swappable protocol cuts both ways, though. The same API that hosts Apple’s models can route your text to outside clouds. The privacy question moves from “which framework” to “which model did this developer pick.”
Where does ILURA stand?
ILURA writes with the on-device Apple Intelligence model. No account, no server, and the App Store privacy label reads Data Not Collected. WWDC26 validates that architecture rather than threatening it.
It also opens a path we are watching. A Private Cloud Compute-hosted model with a 32K context could, in principle, handle much longer documents than the local model manages today — without ILURA adding an account or giving up the Data Not Collected label, because PCC’s privacy guarantees come from the platform, not from us. Whether that holds depends on entitlement fine print we have not finished reading. Treat it as a direction, not a promised feature.
For users today, nothing changes. Playbooks still learn from your corrections, and writing still happens on your phone.
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 did Apple add to the Foundation Models framework at WWDC26?
- Image input, free developer access to a Private Cloud Compute-hosted model with a 32K context window and reasoning, Dynamic Profiles, and a LanguageModel protocol that can route requests to third-party models such as Claude or Gemini.
- Does this mean ILURA will use a cloud model?
- Not today. ILURA runs on the on-device Apple Intelligence model. Private Cloud Compute could one day extend that for long documents without adding accounts, but that depends on entitlement details. It is a direction, not a promise.