LLM in your pocket: An introduction to Foundation Models on iOS

At WWDC25 Apple announced its decision to open the on-device LLM that already powers system features across all platforms to third party developers. Now we can take advantage of a fast, private and on-device LLM using a set of system API provided by the new Foundation Models framework. In this talk we are going to explore the Foundation Models' capabilities to generate contextual text based data in your apps. The main topics will be about setting up the LanguageModelSession, prompting, generating structured data, extend the model's capabilities with Tool calling, handling the response stream and optimizing the speed and efficiency of the LLM. Then we are going to touch on some security best practices to avoid common problems like prompt injection and during the presentation we are also going to show some code snippets that will lead to a final demo app showcasing the capabilities of the on-device LLM.

Speaker Bio:

I've been working as a Software Developer since 2020. During my career, I've worked on many iOS and web projects. Over the past three years, I focused exclusively in iOS development, contributing to three applications that rank among the top 100 in the Italian App Store.