What are all the things the foundation model framework can do?

The Foundation Models framework introduced by Apple at WWDC 2025 offers a wide range of capabilities for developers to enhance their apps with on-device intelligence. Here are the key features and functionalities of the framework:

1. On-Device Large Language Model: The framework provides access to a highly optimized on-device large language model that powers Apple intelligence. This model is available on macOS, iOS, iPadOS, and VisionOS, ensuring privacy as all data stays on the device.

2. Content Generation: It can be used to generate content, such as creating dialogue for game characters or generating itineraries in travel apps.

3. Text Summarization and Extraction: The framework supports tasks like summarizing text and extracting information, which can be used to enhance features like personalized search suggestions.

4. Tool Calling: This feature allows the model to perform actions beyond text generation by accessing live or personal data, such as weather or calendar events, and even executing code defined in your app.

5. Structured Output with Generable: Developers can use the framework to generate structured responses by marking existing types in their app as generable, allowing for guided generation with natural language guides.

6. Streaming Output: The framework supports streaming output, enabling developers to choose between displaying responses incrementally or as a single response.

7. Stateful Sessions: The framework is built around stateful sessions, allowing for multi-turn interactions and maintaining context throughout a session.

8. Safety and Guardrails: It includes built-in guardrails to prevent inappropriate content generation and allows developers to add their own safety rules.

9. Integration with Xcode: The framework is tightly integrated with Xcode, providing a seamless development experience.

10. Adapter Training Toolkit: For specialized use cases, developers can train custom adapters using Apple's Adapter Training Toolkit, although this requires retraining as Apple updates the model.

For more detailed information, you can explore sessions like Meet the Foundation Models framework and Deep dive into the Foundation Models framework.