Transcription
Claude just moved onto your iPhone. Anthropic shipped a Swift package that drops Claude inside Apple's Foundation Models framework.
Here is the split. Apple's on-device model handles the typed Swift values and guided generation. Claude takes the hard part. Multi-step reasoning, code generation, web search, data analysis.
Your app runs local when it can and calls Claude when it cannot. One Anthropic API key wires it all up. And it works across every Apple platform, from iPhone to Vision Pro.
Imagine a journaling app spotting patterns across months of entries. On-device models or Claude in the cloud, which runs your app's brain?