Craft app’s latest AI upgrade may change how you use your iPhone forever

Latest Development: Craft, a document management app, has introduced offline AI capabilities through local language models, marking a significant shift in how AI can be used on Apple devices.

Core Features: The update introduces a flexible AI assistant that allows users to choose between different AI models, including both cloud-based and local options.

• The initial release includes DeepSeek’s R1 model, which requires less than 1GB of storage and uses 1.5 billion parameters
• A more capable 7 billion parameter R1 model is in development, requiring 4GB of storage
• Two Meta LLaMa models are being added, with 1 billion and 3 billion parameters, using less than 700MB and 2GB respectively

Technical Implementation: The local AI models operate completely offline after installation, providing enhanced privacy and accessibility for users.

• All processing occurs directly on the user’s device
• No data is transmitted to external servers
• Users must manually download and enable the models
• The system works across iPhone, iPad, Mac, and potentially Apple Vision Pro

User Benefits: The integration of local AI models provides distinct advantages for everyday document management tasks.

• Enables quick note-taking and text summarization without internet connectivity
• Offers basic reasoning capabilities while maintaining data privacy
• Provides instant response times for common tasks
• Allows seamless switching between local and cloud-based models

Strategic Approach: Craft employs a hybrid strategy, comparing their AI implementation to modern vehicle powertrains.

• Local models serve as the equivalent of electric motors for quick, efficient tasks
• Cloud-based models like ChatGPT handle more complex operations
• Users can choose the appropriate “engine” based on their specific needs
• The system maintains compatibility with server-connected AI models like ChatGPT

Future Implications: This development suggests a growing trend toward local AI processing on personal devices, potentially addressing the current limitations of Apple’s native AI capabilities while maintaining user privacy and data security.