iOSiOS · 2026
SCREEN
把截图变成能找回来的记忆：在本机读懂你存下的东西，用你自己的话搜回来。Screenshots as memory you can find again: understood on device, searched in your own words.
问题The problem
截图是「存了但找不回来」的典型：相册按时间排，回忆却按意义排。存到第一千张时，它已经不属于你了。Screenshots are the classic save-and-lose: the library sorts by time, memory sorts by meaning. By the thousandth one it no longer belongs to you.
做法What I did
Vision 负责 OCR、条码、分类与特征指纹，规则引擎抽实体（网址、价格、日期、品牌、主题）；有 Apple Intelligence 时再让 Foundation Models 补结构化提取，没有它也照常工作。全部在本机，没有账号，没有云端。每条搜索结果都会说明它为什么匹配。Vision handles OCR, barcodes, classification and feature prints; a rule engine extracts entities — URLs, prices, dates, brands, topics. Foundation Models add structured extraction when Apple Intelligence is available, and everything works without it. All on device, no account, no cloud. Every result says why it matched.
结果What changed
真正的价值不在识别率，而在「解释」：找得到，还知道为什么找得到。代价是第一次导入要等分析队列跑完——它不假装瞬间完成。The value is not the recognition rate but the explanation: you find it, and you know why. The cost is the first import, which waits for the analysis queue instead of pretending to be instant.