Apple has launched a new development layer called the Foundation Models framework that gives app makers access to on-device large language models powering Apple Intelligence.
With the rollout of iOS 26, iPadOS 26 and macOS 26, developers can now embed intelligent features within their apps in a privacy-first way, without requiring internet access or extra inference costs.
This marks a turning point in how Apple’s artificial intelligence can be woven into third-party apps. Rather than being a closed system, Apple is now letting independent developers call the same underlying AI models that fuel features across its devices. The framework is tightly integrated with Swift, which means developers can invoke the 3 billion-parameter model with just a few lines of code.
Apple emphasizes that all of this happens locally on device, safeguarding user privacy. No data needs to be sent off to external servers, and apps can function even without a network connection. For more advanced tasks, Apple also uses a companion “Private Cloud Compute” model on its servers, but many everyday AI interactions can be handled completely offline.
From health and fitness to education and productivity, a variety of apps have already adopted the framework and launched smart new features.
For example the SmartGym app can transform a user’s plain text description of a workout into a structured routine with sets, reps and rest periods, adaptively suggest changes over time, and generate insightful summaries. Stoic, the journaling app, can craft deeply personalized prompts based on recent entries and even adjust its style if a user expresses stress or poor sleep.
Other notable integrations include Signeasy, which now can analyze documents, highlight key ideas, and allow natural language queries; Agenda, with its “Ask Agenda” assistant to let users interrogate their note library via conversation; and OmniFocus, which now can propose tasks, tags, and steps based on context.
One notable challenge is that Apple’s models are smaller than some of the models from OpenAI, Google or Meta, meaning the focus is less on giant reasoning tasks and more on real-world assistance features. Still, the ability to enable creative interventions in everyday apps is already reshaping expectations of what apps can do.
Want to see more of our stories on Google?
P.S. Want to keep this site truly independent? Support us by buying us a beer, treating us to a coffee, or shopping through Amazon here. Links in this post are affiliate links, so we earn a tiny commission at no charge to you. Thanks for supporting independent Canadian media!
