在我的iPhone上运行的OpenClaw
最近关于人工智能代理的讨论很多，我在探索一个不同的限制：完全在设备上（iPhone）运行。我正在开发一个类似 OpenClaw 的应用程序，完全在 iPhone 上使用苹果的基础模型运行。所有数据都不离开设备，注重隐私。
目前还处于早期阶段，但围绕本地执行的设计带来了一些有趣的决策。我最多可以同时在后台运行 3 个代理，这样用户仍然可以继续无目的地滚动屏幕。如果超过这个数量，手机就会过热，使用体验会变得很差。
如果有人感兴趣，我很乐意分享更多细节。计划将其开源到 GitHub 上。
查看原文
With all the recent discussion around AI agents, I’ve been exploring a different constraint: fully on device(iPhone).
I’m building an OpenClaw style app to run entirely on an iPhone using Apple’s Foundation Models.
No data leaving the device and privacy focused.
Very early, but designing around local only execution has led to some interesting decisions. I am maxing out at 3 agents running simultaneously in the background so you can still doom scroll. More than that and phone overheats and usability is nonexistent.
Happy to share details if anyone’s curious. Planning to make it open source on GitHub.