[iPhone Version of Doubao Mobile Assistant? Developer Demonstrates Using Opus 5.5 to Operate iPhone] Breaking AI News from Dongcha Beating AI: Developer Finn Voorhees demonstrated an AI Agent capable of directly operating an iPhone. He issued a single command, asking the Agent to find the cheapest Ryanair ticket to London for the next day. The Agent then opened Ryanair, filled in the destination, selected the date, checked ticket prices, and finally returned to the chat interface to report the results—all without involving a computer. This Agent runs on Finn's iPhone chat application, PiChat. At its core, it uses his previously open-sourced PiDurableKit, which leverages Apple's JavaScriptCore to run the Pi Agent on the phone, handling model invocation, tool execution, and task management. The video showcases the use of Claude Opus 5.5, with model inference still performed in the cloud. To ensure the Agent continues working after switching to other apps, PiChat employs Apple's background task mechanism and uses SQLite to save execution progress, allowing recovery after interruptions. The most critical cross-app control code has not yet been made public. In the video, when enabling iPhone Control, the system prompts for the lock screen password to activate XCTest UI automation. XCTest is Apple's interface testing framework, capable of simulating clicks, swipes, and text input. The open-source project TouchSynthesis utilizes XCTest's private interfaces to allow the iPhone to generate touch events independently, without requiring real-time computer control. However, Finn has not confirmed whether this solution was used. [Original Link]