APPSO published a first-hand account of pairing a 21,499-yuan iPhone Duo with an 86,749-yuan Mac Studio to build what it calls a personal AI hub, with the desktop left at home to run models and execute agent tasks while the phone connects remotely. The team wrote a custom app tailored to the Duo's three states: closed, half-folded hover and fully unfolded, giving each a different agent interface — an approval dashboard, a live execution view with voice dispatch, and a two-column console for previewing outputs. On the network side, an HTTPS proxy on the Mac listens on its LAN address and forwards phone requests to a local Bridge service, with Tailscale suggested for cellular access; tasks are tracked by session ID, and a dropped phone connection does not cancel them while the Mac keeps running. Models including DeepSeek V4 Flash, Qwen3.8 27B and GLM 5.3 Flash were served locally through LM Studio, with DeepSeek Harness handling planning and tool use; APPSO said one pelican-on-a-bicycle generation took 2 minutes 18 seconds, averaging 34 tokens per second after two rounds of edits.