Apple has introduced the Foundation Models framework, a native Swift API designed to streamline how developers integrate large language models (LLMs) into their applications. By providing a unified interface, the framework allows developers to swap between local on-device models and remote server-side models without rewriting core logic, significantly lowering the barrier to building complex, multi-model AI experiences.
Unified Access to On-Device and Cloud Models
The Foundation Models framework provides a standardized way to interact with various LLMs. It gives developers direct access to the on-device foundation models that power Apple Intelligence, which are ideal for tasks requiring high privacy and offline functionality.
This image illustrates the core concept of the framework's unified interface for model integration.
For more complex workflows, the framework extends to server-side models including Claude and Gemini. This is achieved through a 'Language Model' protocol, a fundamental component that allows any provider to create a Swift package to join the ecosystem.
“The Foundation Models framework is the best way to use any large language model in your app.”
— Foundation Models framework - Apple Developer
Multimodal Capabilities and Dynamic Profiles
The framework integrates the Vision framework directly into its image understanding capabilities. This allows apps to perform tasks like OCR for text extraction or barcode scanning by simply attaching an image to a text prompt.
This visual supports the explanation of how the framework handles image understanding and multimodal prompts.
A key innovation is 'Dynamic Profiles,' which are declarative APIs for updating a session's model, tools, and instructions on the fly. Because the session transcript is shared across profile switches, apps can maintain context while changing the underlying reasoning level or toolset.
Developer Tools and Cost Considerations
To assist with testing, Apple included an Evaluations framework to help developers 'hill-climb' and refine prompts. This tool is designed to catch behavioral inconsistencies that standard unit tests might miss when switching models or tuning profiles.
While the framework offers broad access, there are specific cost structures for cloud usage. Developers enrolled in the App Store Small Business Program with fewer than two million first-time downloads can access Apple Foundation Models on Private Cloud Compute with no cloud API cost.
This image relates to the framework's ability to utilize Private Cloud Compute for server-side model execution.