What is PhoneClaw?
PhoneClaw is a mobile-native local AI Agent and Agent framework for iPhone. It runs on-device models and native iOS Skills on the phone by default, with optional Mac Gateway remote inference for heavier local models.
Which PhoneClaw project is this?
This project is PhoneClaw for iPhone, an iOS local AI agent and mobile Agent framework.
How does PhoneClaw handle chats, images, and personal data?
Chat, images, and personal data stay on iPhone by default. Web Search, opening a URL, and paired Mac remote inference are explicit user-triggered capabilities.
How does PhoneClaw work as a local iPhone Agent?
PhoneClaw uses local iPhone models, native iOS Skills, permission boundaries, fully offline local execution, and user-triggered network capabilities for Web Search, URL reading, and Mac remote inference.
How does PhoneClaw work as a mobile-native Agent framework?
PhoneClaw is designed for mobile and edge-device constraints: iOS permissions, app lifecycle limits, memory budgets, model downloads, widgets, Shortcuts, Dynamic Island, voice, camera, and optional LAN Mac inference.
What models does PhoneClaw use?
PhoneClaw uses Gemma 4 E2B and E4B via LiteRT-LM for local language tasks, and MiniCPM-V 4.6 for image understanding and LIVE camera scenarios.
What native iOS Skills can PhoneClaw use?
PhoneClaw Skills can work with native iOS capabilities such as Calendar, Reminders, Contacts, Clipboard, HealthKit, translation, images, and explicit Web Search, subject to iOS permissions and Skill allowlists.
What is PhoneClaw Gateway?
PhoneClaw Gateway pairs an iPhone with a Mac on the same local network. The Mac can provide heavier inference through Ollama or supported CLI providers, while the iPhone keeps the agent UI, Skills, permissions, and task flow.