# 短信滤盒 / FilterBox

短信滤盒是一款 iPhone 离线 AI 短信过滤应用，使用 Apple Core ML 在设备上分类短信，并从用户主动投喂的内容中学习个人偏好。繁体中文名称为「簡訊濾盒」，英文和日文名称为 FilterBox。它是 Android「通知滤盒」的 iOS 版，iOS 版聚焦短信过滤。

## 官方入口

- [官网](./)
- [App Store](https://apps.apple.com/cn/app/id6815764571)
- [使用支持](./support.html)
- [隐私政策](./privacy.html)
- 联系邮箱：privacy@catchingnow.com
- 开发者：Ruoxin He

## 适用设备与语言

- iPhone，iOS 27 或更高版本。
- 界面支持简体中文、繁体中文、英语和日语。
- 短信识别目前以中文短信为主。界面语言支持不等于所有语言具有相同识别效果。

## 核心功能

### 自动短信过滤

在系统「信息」设置中选择短信滤盒作为短信过滤器后，过滤扩展会将 iOS 交给它的短信分类为正常或营销。正常短信包括验证码、物流、支付和预约通知；广告、推销、骚扰等归入营销。实际分类结果在系统「信息」中查看。

### 个人投喂与学习

用户可以复制短信，主动粘贴到 App，选择「是骚扰」或「是正常」，不用编写过滤规则。解锁个人学习后，相同模板立即使用用户的分类。系统在充电时安排个人模型学习，帮助识别相近内容；后台学习不保证立即完成。

用户可以查看学习记录、修改分类或删除投喂。删除全部投喂，也会移除个人学习结果。

### 算法核心

长按 App 内跳动的火种，可打开「算法核心」，观察短信内容、判断与学习过程的可视化。官网算法演示使用 App 内置 CoreSamples 样例，概率与特征哈希由 App 分类器在构建时预计算。投喂界面使用已有宣传素材样例。网页不包含或运行短信模型，不接受或上传用户短信。

## 隐私

短信会上传吗？不会！分类和个人学习都在 iPhone 本地完成，无需注册账号，不向开发者服务器上传短信、投喂记录或个人模型。App 无法浏览收件箱。只有用户主动粘贴的短信会进入学习库。

App 不提供开发者云端同步。系统备份是否包含 App 数据取决于设备备份设置。购买和恢复购买由 Apple 处理。官网由 Cloudflare 托管，不嵌入广告或分析追踪脚本，托管服务的请求数据处理见隐私政策。

## 付费方式

基础过滤免费。「本地 AI 自进化功能」为一次性非消耗型内购，无订阅，价格以用户所在地区的 App Store 为准。

免费用户可在 App 的「算法核心」体验投喂与学习，数据保存在本机；解锁前，个人学习结果不参与日常短信过滤。已经购买的用户可以恢复购买。

## 系统边界

- 过滤范围由 iOS 决定，联系人、已知发件人及 iMessage 不在本扩展的过滤范围内。
- App 无法读取收件箱，也无法展示逐条拦截历史。实际短信分类请在系统「信息」中查看。
- 学习有助于适应个人偏好，不承诺所有短信都能被正确分类。

## English product facts

FilterBox is an iPhone app for offline SMS filtering and personal learning with Apple Core ML. It requires iOS 27 or later. The interface supports English, Simplified Chinese, Traditional Chinese, and Japanese; SMS recognition currently focuses on Chinese.

After setup in system Messages settings, FilterBox classifies eligible SMS as normal or promotional. Users can copy a message, paste it into the app, and label it Spam or Normal. Once personal learning is unlocked, matching templates use the chosen label immediately. iOS schedules model learning while charging to help with similar messages; completion is not guaranteed to be immediate.

Are texts uploaded? No! Filtering and learning take place on the iPhone. No account is required, and texts, teaching examples, and personal models are not uploaded to developer servers. The app cannot browse the inbox. Only voluntarily pasted texts enter learning history. Examples can be reviewed, relabeled, or deleted; deleting all examples also removes personal learning results.

Basic filtering is free. Local AI self-learning is a one-time, non-consumable in-app purchase, with no subscription. Users can try teaching in Algorithm Core before purchase; personal learning is applied to everyday filtering after unlocking it. Regional pricing is shown in the App Store, and purchases can be restored.

Press and hold the app's glowing ember to open Algorithm Core. The website uses the app’s built-in CoreSamples, with probabilities and feature hashes computed by the app’s classifier at build time. It ships only the samples and results, without the model. iOS determines filtering eligibility; contacts, known senders, and iMessage are outside this extension's scope. Actual results are viewed in Messages.