知在 Awareful 隐私政策 · Privacy Policy
在中国大陆 App Store 创建的账号（中国大陆区）适用下面的中文隐私政策；在其他地区 App Store 创建的账号（海外区）适用 English 部分。
Accounts created from the mainland-China App Store (China mainland region) are covered by the Chinese text below; accounts created from any other App Store storefront (Global region) are covered by the English Privacy Policy.
中文（中国大陆区）
更新日期： 2026 年 10 月 30 日 生效日期： 2026 年 10 月 30 日
运营者： 搜达库（海南）科技有限公司（下称「我们」）
注册地址： 海南省海口市龙华区大同街道龙华路17号财盛大厦第16层C区C900号
个人信息保护联系方式： privacy@awareful.com
个人信息保护负责人： 可通过 privacy@awareful.com 联系
ICP 备案： 知在 App 备案号 琼ICP备2025061517号-11A
知在（英文名 Awareful）让你在 iPhone 上监督 Mac 上的 Claude Code、Codex、Cursor 等 Agent：它们真需要你时，审批和提问送到手机，你的答复回到原来的会话；也可以从手机给 Agent 发话、开新任务。本政策说明你使用知在时，我们处理哪些个人信息、为什么、放在哪里、留多久，以及你如何管理它们。
本政策适用于：iPhone 上的知在 App；Mac 上的 Awareful Desk 及它为你装进 Agent 的连接程序；我们的中继与内容服务；你发布成链接的公开页面；官网。第三方的产品和服务按它们自己的政策处理，见第三节。
请特别留意（加粗部分与你的权益关系较大）：
- 数据存放在中国境内。 中国大陆区账号的数据存放在阿里云上海。区域按你创建账号时的 App Store 商店地区确定，不随你所在地点或旅行改变。
- 端到端加密与服务器可读，分两层。 会话镜像、Agent 原生弹出的审批和提问的内容、你从手机发给 Agent 的话、附件和成果，在你的 Mac 与配对的手机之间端到端加密，我们的服务器只见密文。你的任务、子任务、记录、想法、Agent 经知在问你的问题和你的回答、留言等，保存在我们的服务器上，我们的服务器能够读取。
- Agent Use 可能把屏幕截图送给模型。 你另行同意「模型辅助」（Awareful Desk 替你操作 Agent 应用时，脚本走不通可请模型帮忙；默认关闭，各档位一样）后，脚本走不通时，Desk 可能先把目标窗口的控件列表、再必要时把目标 App 窗口的截图（连同盖在这些窗口上的系统对话框）经我们的服务送到阿里云百炼的模型。截图里可能有窗口里显示的任何内容。我们自己不保存截图；百炼会按法律法规留存模型调用数据。 见 1.5 与 3.2。
- 简报不经模型整理。 回来简报按规则从真实发生的事件算出，不把你的记录交给模型整理。
- 你在 Mac 上让 Claude Code、Codex、Cursor 等接入知在后，它们读到的记录会被它们交给各自的模型服务（多数在美国）处理，见 3.4.2。
- 中国大陆区目前不开放远程连接（例如在 ChatGPT 里连接知在），我们不会把你的记录提供给其他 App。见 3.4.1。
- 知在不接入第三方广告或统计工具，不做用户画像，不做个性化推荐，不出售你的个人信息；App 没有嵌入第三方 SDK（只用 Apple 的系统服务）。除实现基本功能所必需的信息外，你可以拒绝提供；相机、通知、Face ID 等权限只在用到相应功能时申请，可随时在系统设置里关闭。
一、我们如何收集和使用个人信息
1.1 注册与登录
| 方式 | 说明 |
|---|---|
| 手机号验证码 | 提供中国大陆手机号。号码明文只在这一次发送请求里用来交给阿里云短信服务发验证码，我们不保存号码明文；保存的是用密钥运算后的值（用来下次认出同一个号码）和遮罩后的显示标签（如 +86138****1234）。验证码 5 分钟内有效。 |
| 通过 Apple 登录 | 我们收到经签名的身份凭证，只保存 Apple 给出的用户标识，不保存你的邮箱和姓名，也不保存访问令牌。 |
为确定账号区域，我们读取 App Store 商店地区代码；为在设备列表里区分设备，我们记录设备型号名称（如「iPhone 17 Pro」）。为防止验证码被滥用，我们记录号码、IP 地址、App 生成的随机设备标识各自经运算后的值和发送时间，24 小时后删除，不保存原始号码或 IP。几种方式可以绑定到同一个账号，也可以解绑，但至少保留一种。不提供上述信息，你将无法注册和登录。
1.2 配对、会话镜像、审批与发送
- 配对： 你用手机扫描 Awareful Desk 上的配对码，我们申请相机权限，只用于扫码；第一次配对新手机时，你要在 Mac 上核对 4 位数字并确认。手机和 Mac 各自生成并交换用于加密和签名的公钥；中继登记设备标识、公钥、配对关系、推送令牌和最近一次连接的时间；手机的名称只在等你在 Mac 上确认配对期间暂存。中继不保存 IP 地址。
- 会话镜像： Awareful Desk 读取你在 Mac 上选定范围内的 Claude Code、Codex、Cursor 会话，在 Mac 上加密后经中继发到你配对的手机，只有你的 Mac 和配对的手机能解开。
- 审批与提问： Agent 自己弹出的审批和提问，内容端到端加密送到手机；我们的服务器只保存请求编号、所属任务、来源会话、是否只能在 Mac 上处理、过期时间和处理状态（何时由谁答复、何时送达），不保存命令、题目和答案的内容。你的答复用只保存在你 iPhone 上的私钥签名；需要你本人确认的操作使用 Face ID。
- 发送与附件： 你从手机发给 Agent 的话、照片和文件，以及 Agent 交给手机的成果，都端到端加密传输；我们只保存密文，取到即删。选照片用系统照片选择器，我们只取得你选中的那几张。
1.3 Awareful Desk 在你的 Mac 上替你操作
你让知在给某个会话发话、新建或 fork 会话、修改会话设置，或答复一组原生提问时，Awareful Desk 在你的 Mac 上、在目标 App 自己的窗口里替你完成：优先使用 App 的辅助功能控件，必要时对该窗口做定向的后台输入，不移动你的鼠标。为此需要你在 macOS 授予辅助功能和屏幕录制权限；屏幕录制用于看清目标窗口、确认操作落实，你离开时 Desk 也用它遮住屏幕（不锁屏）。
- 每次操作的运行记录（阶段、结局、诊断用的原始事件）和操作中看到的画面只保存在你的 Mac 上，不上传；唯一的例外是 1.5 和 1.6 所述。
- 回给手机的回执只有阶段和结局，不含消息正文，端到端加密。
- Awareful Desk 会向我们报告：这台 Mac 是否遮屏或锁屏、何时离开与回来、Agent 会话所在工作目录与 git 仓库的路径（只有路径，不含代码）、原生审批和提问的元数据。
- Desk 用来认出 Agent 应用控件的描述随 Desk 一起安装，不从我们这里下载。
- 你可以随时在 macOS「系统设置 › 隐私与安全性」收回权限。
1.4 你的任务与记录（服务器可读）
这是知在的主体功能，以下内容保存在我们的服务器上，我们能够读取：
- 任务与子任务： 标题、你的原话、做完是指什么、状态、提醒、任务之间的前后关系；归你自己的待办同样保存在这里。
- 记录条目： 你或 Agent 记下的决定、改了方向、结果、风险、交接、Agent 发现等文字，及其类型、状态和修订历史。
- 想法、灵感与书签： 你记下的想法和灵感，书签里你写的话；书签附带的会话摘录在你的设备上加密，我们只存密文，不能解开或搜索。
- 请求： Agent 经知在问你的问题、选项和你的回答，以明文保存。
- 留言、偏好、Skill 库、Agent 提议与反馈；领域和整理建议及你的采用、不用和撤销。
- 来源： 每条内容是哪个 Agent 在哪个会话里写的（会话标识和标题）。
- 你的动作与焦点： 你最后一次对某个任务做了什么、何时（附你的原话或短名）；你当前关注的任务和会话；离开与回来的时间和来源。
- 模式： Build、Wild、Sleep 的当前模式、切换时间与来源、起床时间、时区和提醒设置。
- 工作树路径： 会话所在的工作树和主 git 仓库的路径（只有路径，不含代码），用来整理建议。
- 头像（可选）： 你选的一张小图（不超过 128 KB，不带位置等元数据）。
- 阅读位置： 你上次看完简报的时间，以及动态在各个范围（全部、某件事、某台 Mac 上的某个 Agent）各读到的时间，最多 200 个范围；只有编号和时间，不含内容。
- Agent 的读写日志： 你接入的 Agent 每次调用知在，我们记下调用的工具、参数的形状和其中的搜索词、返回了多少条、涉及的条目编号、经核实的会话、耗时和结果，不记条目正文；保存 30 天，你可以在「我 › Agent 用了什么」里查看。
回来简报与引用的原话。 你从 Wild 或 Sleep 回到 Build 时，我们按规则从真实发生的事件冻结一份简报：它可以包含你对任务说的原话、你记的灵感、问题与答案、留言，以及从镜像会话里引用的短短几句原话（例如你走时最后说的那句、你在现场发的话）。这些引用的原话以明文存放在我们的服务器上，并可供你搜索；我们不这样存整段会话，原生审批和提问的内容与答案也不进简报。撤销那次切回，或删除原话所属的任务或灵感，对应文字随之清除。简报由规则生成，不经模型整理。
请注意： 记录里写什么，由你和你的 Agent 决定。如果写入了你自己或他人的敏感个人信息（例如健康、金融账户、行踪），它们会和其他内容一样被保存、同步，并可能按 3.4 交给你接入的 Agent 或 App。请不要在记录里写入密码和密钥；写入他人个人信息前，请确认你有权这样做。
1.5 Agent Use：脚本路径与模型路径
从手机使用 Agent Use 随 Plus 和 Pro 档提供，新账号的前 14 天 Plus 期间同样可用。下面的数据处理对所有账号一样，包括 Free 账号用到模型路径的时候；文中「模型辅助」指任何账号的模型路径。Agent Use 按顺序用三层：目标 App 自己的原生通道、已验证的脚本、模型。
脚本路径（绝大多数操作）： 全部在你的 Mac 上完成，不涉及模型。
模型路径（慢路径）： 只有脚本走不通时，Desk 才可能请模型决定下一步，分两步：
- 先把目标窗口里可见控件的列表（控件的角色、名称、位置）送去，会话标题和草稿内容在列表里换成占位符；
- 列表不够用时，才送去目标 Agent 应用在那块屏幕上的窗口截图，包括盖在这些窗口上的系统对话框（例如选择文件的面板、权限提示）；其他 App 的普通窗口和知在自己的窗口不在截图里。截图里可能有这些窗口和对话框显示的任何内容：对话、代码、文件名、文件夹路径、其他 App 的提示、个人信息、密钥。 Awareful Desk 不能选择窗口里显示什么；如果窗口里有你不愿让模型服务处理的内容，请不要在那时使用模型路径。
传输路径（中国大陆区）： 你的 Mac → 我们在阿里云上海的内容服务 → 阿里云百炼（由阿里云计算有限公司提供）的模型。全程在中国境内。
我们保留什么： 我们不把截图和控件列表写入磁盘或数据库。 我们为执行用量上限记下每次模型调用：属于哪次运行、宿主 App、第几步、模型、token 数、费用、耗时、列出的控件个数和截图的字节数；不记控件内容、截图、任务和模型的回答。这些记录保存 90 天，账号删除时一起删除。
百炼一侧： 阿里云百炼的公开说明称，不会把你的数据用于模型训练，同时按法律法规要求存储模型与应用调用时产生的数据（阿里云帮助中心《合规资质与隐私说明》）；具体留存期限以《阿里云百炼服务协议》和阿里云的隐私政策为准。
保护措施： 模型路径只做可撤回的动作（找会话、滚动、关菜单、聚焦输入框、写入正文，写了没发可以清掉）；不可撤回的发送之前，Desk 会先确定性地核对会话、整段正文和附件；完成只认目标 App 自己的回执。
同意与关闭： 这里有两个互相独立的控制。（1）「Agent Use 开关」（Desk 和手机上的 Mac 设置里）让手机上的发话、新建会话、改设置经脚本路径在 Agent 应用里生效，它本身不把任何内容送给模型。（2）「模型辅助」是另一项选择，默认关闭，要在单独的界面里同意；这个界面写明接收方（我们的服务、阿里云百炼）、截图可能含窗口里的任何内容、百炼会按法律法规留存模型调用数据，且不与其他同意捆绑，你不同意不影响其他功能。没有同意（2）时，Desk 只用目标 App 自己的原生通道和已验证的脚本，走不通就把事情交还给你。我们按账号记下你同意的是哪一版文字和同意的时间；没有同意当前这一版时，我们的服务在调用模型之前就拒绝。（2）可以随时在 Agent Use 开关旁撤回。
自动化决策： 模型只是在你自己电脑上的应用里选择点击位置，不对你个人作出对你权益有重大影响的决定，也不用于个性化推荐。
1.6 失败报告（Agent Use）
这台 Mac 开着失败报告时，操作出问题或脚本交给模型（脚本放手、发送前核对没通过、模型路径结束、结局是「尚未确认」或「需要人处理」），Desk 向我们发一份小报告，我们据此修适配器。纯脚本的一次成功不单独报告，只随每天一条汇总报次数。
字段： 运行编号与时间；宿主（codex、claude-code、cursor）、宿主 App 的版本和构建号、界面语言；Desk、内核、适配数据版本；操作类型；走过的路径（App 自己的通道、脚本、模型）；脚本为什么放手（原因代码，如「没找到输入框」）与在哪一步；模型用量（决策次数、截图张数、费用、供应商、模型）；哪一项发送前核对没过；结局与原因代码；适配器期望的每个控件是「找到／缺／有多个」。
不包含： 消息正文、标题、会话编号、截图、原始辅助功能标签、模型的思考内容、文件名。
保存 90 天，只用于定位和修复适配问题（运维人员按 App 版本看次数），账号删除时一并删除。默认开启（这台 Mac 在 macOS「系统设置 › 通用 › 语言与地区」里的地区是欧洲经济区或英国时默认关闭），每台 Mac 可以单独开关：在 Awareful Desk 的「隐私」页（Agent Use 一行下的「失败报告」），或在手机上这台 Mac 的卡片里点「关闭报告」（关着时这个按钮是「打开报告」）。你的选择优先于默认。关着时这台 Mac 不发送；关闭时还没发出的报告也一并丢弃。
1.7 通知、实时活动与小组件
你允许通知后，我们把推送令牌登记到中继和内容服务，经 Apple 推送通知服务（APNs）发送：推送里只有编号、固定文字和数字（如请求编号、「有事需要你处理」），不含会话内容、题目、答案或任务名，具体内容由手机上的知在解开或从服务器取回后显示。锁屏与灵动岛的实时活动经 Apple 推送时，只带「需要你」「进行中」「今天已完成」「需关注」的数量、Desk 状态、开始遮屏的时间、是否已配对，以及最近一步的种类和时间（完成、卡住、提问、推进一步），不带任何标题或文字。手机收到后会在本机补上任务短名、会话标题和请求在做什么（例如「需要你 1 · 做原型 C · Claude 要运行命令」），看得到你锁屏的人都能看到这些文字。主屏小组件读的是手机上的数据，不经 Apple 推送。手机会报告时区，以便在当地时间提醒你。
1.8 发布成链接
你可以把成果发布成公开链接；Agent 替你发布，必须带你在这次对话里的原话，或在手机上得到你的「允许」。我们处理：发布的文件（明文放在你账号所在区单独的存储桶）、标题、版本、来源会话与任务，以及你的同意记录。拿到链接的任何人都能看，链接里的随机地址不代表隐私；下架后立即失效，但已被他人看到或保存的内容我们无法收回。每个版本上线前，其中的文字和图片会交给阿里云内容安全（上海）审核，我们只保存审核回执（请求编号、风险级别、标签）。公开页面不设置 Cookie，我们不记录访客的 IP 或身份，按站点只记当天的打开总数；另按账号记每月这些链接送出的字节数和请求次数，用于流量上限，不含任何访客信息。
1.9 档位、订阅与用量
我们记录你在手机上每次操作的种类和时间（批准、回答、发送、新建会话）和当时的档位，不含内容，用来确定权益和防止滥用。新账号自配对 Mac 起有 14 天 Plus，我们记下这段的起止时间。订阅由 Apple 处理，我们接触不到你的银行卡和账单地址；我们从 Apple 收到经签名的交易信息（商品、交易编号、原始交易编号、环境、商店地区、订阅状态与到期时间）来确定权益。
1.10 远程连接
中国大陆区目前不开放远程连接（例如在 ChatGPT 里连接知在），我们不为此处理你的任何信息。开放之前，我们会更新本政策，并在你连接前单独告知接收方和出境事项、取得你的单独同意。
1.11 安全保障、诊断与客服
为保障服务安全、防止滥用和攻击，我们处理你访问服务时的 IP 地址（用于限流的部分只存运算后的值）、请求时间和结果。服务端日志只记服务事件和错误（例如设备取消息的情况、推送失败、服务出错，含设备标识和客户端版本），不记录记录正文和密钥，网站服务器不记访问日志；服务端日志保存不少于六个月（见 4.3）。iOS 交给知在的崩溃和卡顿报告只留在你的手机上；只有你在 App 里导出诊断包并发给我们，我们才会收到，诊断包不含私钥、令牌、会话正文和你写的话。你写信给我们，我们处理你的邮箱地址和你描述的情况，用来回复。
1.12 演示模式
App 里有演示模式，方便审核人员和你自己看看、给朋友展示。演示模式使用打包在 App 里的虚构示例数据，放在单独的存储里，读不到也改不了你真实账号的数据，不需要账号和 Mac。App 的请求在手机上由这些示例数据直接回答，其他请求一律拒绝，不经网络发出；你在演示模式里的操作不会发送给我们，也不会用到你的 iCloud、购买和通知。
1.13 只在你设备上的数据
手机上有一个本地全文索引，供你搜索镜像会话里人说的话和 Agent 的正文（不含工具调用、输出和思考），不上传，随镜像同删。答过的原生题及答案、动态上的「赞」只存在手机上。成果与附件缓存上限 500 MB，可在「我 › 存储」清除。你在 App 里读到哪儿也在你的设备上记一份，并经你自己的 iCloud 键值存储在你的设备之间同步，我们读不到；我们的服务器上也记着阅读位置，见 1.4。
1.14 我们不收集的信息
我们不收集你的精确位置、通讯录、广告标识符和浏览记录，也不读取你没有选中的照片。会话整段内容和你的代码不会以我们能读取的形式上传到服务器（1.4 所述的简报引用原话除外）。
1.15 依法无需征得同意的情形
根据法律法规，下列情形我们处理你的个人信息无需取得你的同意：为订立、履行你作为一方当事人的合同所必需；为履行法定职责或者法定义务所必需；为应对突发公共卫生事件，或者紧急情况下为保护自然人的生命健康和财产安全所必需；为公共利益实施新闻报道、舆论监督等行为，在合理的范围内处理个人信息；依照法律规定在合理的范围内处理你自行公开或者其他已经合法公开的个人信息；法律、行政法规规定的其他情形。
二、Cookie 和同类技术
知在 App 不使用 Cookie。App 在你的设备上保存登录状态、加密密钥（存于系统钥匙串）、设置和缓存；退出登录时删除。官网是静态网站，不使用 Cookie，不接入统计或广告工具；你发布的公开页面也不设置 Cookie。
三、合作方，以及委托处理、提供、转移、公开个人信息
3.1 原则
与合作方合作涉及你的个人信息时，我们遵循合法、正当、必要，以达成目的所必需为限，并审慎评估合作方的目的和安全保障能力。
3.2 委托处理
部分功能由合作方受我们委托提供：云服务器与数据库、边缘加速、短信发送、内容审核，以及 Agent Use 的模型处理（阿里云百炼）。我们与受托方约定处理的目的、期限、方式、个人信息的种类和保护措施，并监督其处理；受托方只能在约定范围内处理。名单见附录二。
3.3 向第三方提供
除下列情形外，我们不会向其他个人信息处理者提供你的个人信息：为实现登录、通知和 App 内购买，Apple 依其自身规则处理相应信息（附录二）；你主动发布成公开链接（1.8）；1.15 所列情形。
3.4 你接入的 Agent 与 App
3.4.1 远程连接的 App（例如 ChatGPT）。 中国大陆区目前不开放，见 1.10。
3.4.2 你在 Mac 上运行的 Agent。 你在 Awareful Desk「接入 Agent」里让 Claude Code、Codex、Cursor 等接入知在后，Desk 在你的 Mac 上为它们装好连接程序，它们工作时通过连接程序读写你的任务与记录。这些 Agent 是你在自己 Mac 上、用你自己的账号运行的第三方软件，它们读到的记录会连同你的其他对话内容一起交给各自的模型服务：Claude Code 交给 Anthropic，Codex 交给 OpenAI，Cursor 交给 Anysphere 并由它转给你选的模型服务。这些模型服务多数位于美国，按各自的隐私政策处理。运营者和联系方式见附录三。不希望某个 Agent 读取你的记录，可以在「接入 Agent」里对它点「移除…」。
3.4.3 第三方自己的处理。 对方 App 和 Agent 拿到信息之后如何处理，由它们按各自的条款和隐私政策负责，不适用本政策。
3.5 转移
因合并、收购、资产转让等需要转移你的个人信息时，我们会告知你接收方的名称和联系方式，并要求接收方继续履行本政策下的义务；接收方变更原先的处理目的、处理方式的，会重新取得你的同意。
3.6 公开
我们不会公开你的个人信息，除非你主动发布成公开链接（1.8）、法律法规另有规定，或取得你的单独同意。
四、我们如何存储个人信息
4.1 存储地点
中国大陆区的数据存放在中国境内：阿里云（上海）的服务器、数据库与对象存储；发布的链接放在上海单独的存储桶，经阿里云边缘安全加速（ESA）以 <链接名>.awareful.me 提供。海外区与中国大陆区的设施互相独立，不共享账号、不互相备份、不互相调用。
4.2 个人信息出境
我们把在中国境内运营中收集和产生的个人信息存储在中国境内。Agent Use 的模型处理（阿里云百炼）发生在境内，不出境。 除下列情形外，我们不会向境外提供你的个人信息：
- 你在 Mac 上运行的 Agent 把读到的记录交给它的模型服务时。 这是你自己运行的第三方软件把信息发往境外，见 3.4.2。
- 发送通知、通过 Apple 登录和订阅结算时。 经 Apple 的服务（Apple Inc.，美国），只涉及推送令牌、身份凭证、交易信息和 1.7 所述的推送内容，不含会话正文和记录正文。
- 你发布的公开链接被境外访客打开时。 公开内容按你的选择对任何人可见。
向境外提供你的个人信息前，我们会依法开展个人信息保护影响评估，并按法律、行政法规和国家网信部门的规定，在需要时通过安全评估、个人信息保护认证或与境外接收方订立标准合同等方式满足条件。
4.3 存储期限
| 信息 | 保留期限 |
|---|---|
| 任务、记录条目、想法、书签、领域 | 保留到你删除。删除后进入「最近删除」，30 天内可恢复，之后连同修订彻底删除 |
| 待办与子任务 | 保留到你删除，删除即彻底删除 |
| 请求（问题与回答、原生请求元数据）、留言、交接、偏好、Skill、Agent 反馈、你的动作与焦点、模式与离开记录、简报（含引用的原话）、简报与动态的阅读位置、头像、操作计数、发布链接的每月流量计数 | 随账号保留，账号删除时一起删除；未被取走的领域留言 7 天后过期 |
| 整理建议与你的采用记录 | 保留到账号删除；采用后 30 天内可撤销 |
| 会话镜像、审批、答复的密文（中继） | 所有收件设备确认收到即删；没送达的镜像最长 30 天，审批和答复最长 1 天 |
| 附件与成果的密文 | 所有收件设备取到即删；最长 7 天；登记后 24 小时没传完的删除 |
| 你 Mac 上的附件与成果 | 手机发来的附件 7 天；成果最新版本保留到你删除，更早的版本 90 天 |
| 手机上已结束的审批、已完成的发送记录 | 7 天 |
| 发布的链接 | 当前版本保留到你下架；其他版本 90 天；下架后 7 天内可恢复，之后删除站点、版本和文件；审核回执随版本删除；每日打开数只留当天 |
| Agent 和应用读写记录的日志（1.4 所述） | 30 天 |
| 写操作的防重复回执 | 30 天 |
| 服务端运行日志（服务事件与错误，例如设备取消息、推送失败、服务出错，含时间、设备标识和客户端版本；不含正文和密钥；网站服务器不记访问日志） | 不少于六个月（《网络安全法》要求网络日志留存不少于六个月），之后删除 |
| Agent Use 失败报告 | 90 天 |
| Agent Use 截图与控件列表 | 我们不保存（只在处理一次请求期间存在；百炼一侧按其协议处理，见 1.5） |
| Agent Use 模型调用记录 | 90 天；账号删除时一起删除 |
| Agent Use 同意记录（同意或撤回、文字版本、时间、手机或 Desk） | 随账号保留，账号删除时一起删除 |
| 每台 Mac 每天经脚本成功的 Agent Use 次数（宿主、App 版本、次数，不含内容） | 90 天 |
| 「Plus 天数」记录（前 14 天 Plus 的起止时间） | 随账号保留，到期不删，账号删除时一起删除 |
| 登录方式 | 解绑时立即删除；账号删除时一起删除 |
| 验证码与授权流程 | 短信验证码 5 分钟、通过 Apple 登录的流程 10 分钟内有效，过期后 1 小时内清理；防滥用计数 24 小时 |
| 推送令牌 | 设备移除、退出登录、账号删除或 Apple 告知失效时删除 |
| 数据库备份 | 7 天滚动；已删除的数据最晚在删除 7 天后不再出现在任何备份中 |
你 Mac 和手机上的数据随卸载 App 删除。超出期限后，我们会删除或匿名化你的个人信息，法律法规另有规定的除外。
五、我们如何保护个人信息的安全
- 传输： 设备与服务器之间全程 TLS 加密。
- 端到端加密： 会话镜像、原生审批与提问、答复、发送、附件与成果在你的设备上加密，只有你的 Mac 和配对的手机能解开；中继和对象存储只存密文。
- 签名与本人确认： 批准和答复用只保存在你 iPhone 上的私钥签名；需要你本人确认的操作使用 Face ID；第一次配对新手机要在 Mac 上核对数字并确认。
- 最少保存： 手机号、IP 只保存运算后的值；令牌只保存摘要；服务端日志不记录正文和密钥。
- 隔离： 两区设施各自独立；发布的公开页面在隔离的沙盒里运行，读不到账号数据和 Cookie。
- 安全事件： 如发生个人信息安全事件，我们会立即启动应急预案、采取补救措施，并按法律法规要求及时通过推送通知、邮件等方式告知你事件的基本情况、可能的影响、处置措施和你可以采取的防范建议；难以逐一告知时发布公告，并按监管要求报告。
六、管理你的个人信息
你可以在 App 里直接完成，无需联系我们：
- 查阅、复制和转移： 任务、记录、待办都显示在 App 里；「我 › 账号 › 导出我的记录」导出一份 JSON（领域、任务、条目及修订、子任务、请求、留言、偏好与设置、会话、离开记录、Skill、停点与焦点、引用、发送与交接、头像；书签的加密摘录按密文导出）。
- 更正和补充： 可修改你写的任务、记录和待办；对 Agent 写的内容有异议，可删除或联系我们。
- 删除： 可删除单条记录、想法、书签（30 天内可在「最近删除」恢复）、待办和头像，下架发布的链接，移除一台 Mac 或手机。删除后备份里的数据最晚 7 天内随备份过期删除。
- 撤回同意： 本机 Agent 在 Desk「接入 Agent」移除；模型辅助和 Agent Use 开关在 Desk 和手机的 Mac 设置里关闭（失败报告见 1.6）；通知、相机、Face ID 在 iPhone「设置 › 知在」关闭；辅助功能、屏幕录制在 macOS「系统设置 › 隐私与安全性」关闭。撤回不影响撤回前已进行的处理。
- 登录方式： 在「我 › 账号 › 登录方式」绑定或解绑，至少保留一种。
6.1 注销账号
在「我 › 账号 › 删除账号」发起，发起后立即生效：所有设备解除配对、登录失效，中继上待投递的内容和推送令牌立即删除，发布的链接立即停止公开访问。7 天内用同一种登录方式重新登录可以撤销；7 天后我们彻底删除账号和全部内容（包括发布的文件）；备份最晚再过 7 天过期，即最晚 14 天后不再有任何副本。注销前可先导出。注销不会取消 App Store 订阅，请先在 Apple 订阅设置里取消。 你 Mac 和手机上的本地数据随卸载 App 删除。
6.2 其他请求与响应
上述方式无法完成的请求（解释本政策、对 Agent 写入内容提出异议等），请按第九节联系我们。为保障安全，我们可能先验证你的身份，并在验证后十五个工作日内回复。对于无端重复、需要过多技术手段、给他人合法权益带来风险或不切实际的请求，我们可能拒绝并告知理由。
6.3 停止运营
如果我们停止运营知在，会停止收集你的个人信息，通过 App 内通知或公告通知你，并在合理期限内删除或匿名化你的个人信息。
七、未成年人
知在面向成年人，主要用于工作。知在不面向不满十四周岁的未成年人。 如果我们发现在未事先取得监护人同意的情况下收集了不满十四周岁未成年人的个人信息，会尽快删除。
八、隐私政策的修订
我们可能适时修订本政策。未经你明确同意，我们不会削减你按本政策享有的权利。修订后我们会在官网和 App 里发布新版本，并在生效前通过 App 内提示等方式告知更新内容。处理目的、方式、种类，存储地点与出境情形，接入的接收方，运营者或你行使权利的方式发生重大变化时，会以显著方式提示，需要你同意的重新取得同意。我们保留历史版本供你查阅。
九、联系我们
知在的运营者是搜达库（海南）科技有限公司，注册地址见本政策开头。对本政策或你的个人信息有任何疑问、意见或请求，请发邮件到 privacy@awareful.com（其他问题可发到 support@awareful.com）。我们会在验证身份后十五个工作日内回复。如果你对回复不满意，可以向网信、电信、公安及市场监管等部门投诉、举报，或向有管辖权的人民法院提起诉讼。
十、其他
本政策只有简体中文版本，适用于中国大陆区账号。同一网页上的英文部分是海外区账号适用的另一份政策，不是本政策的译文。标题仅为方便阅读，不影响条款的含义。
附录一：已收集个人信息清单
| 功能 | 收集的信息 | 目的 | 是否必需 | 存放 |
|---|---|---|---|---|
| 注册登录（手机号） | 手机号（只在发送验证码时用明文，随后只存运算值与遮罩标签）、验证码流程数据、商店地区、设备型号名称、IP（运算值） | 创建和认出账号、防滥用、确定区域 | 必需 | 阿里云上海 |
| 注册登录（Apple） | Apple 用户标识 | 同上 | 必需（二选一） | 阿里云上海 |
| 配对与会话镜像 | 设备标识、公钥、配对关系、推送令牌、最近连接时间、配对确认期间暂存的手机名称；会话镜像密文 | 把 Mac 上的会话送到手机 | 使用核心功能时必需 | 中继（密文） |
| 审批、答复、发送、附件 | 请求编号、所属任务、来源会话、状态、过期时间；密文 | 把审批和答复送达 | 使用核心功能时必需 | 中继／对象存储（密文） |
| 相机 | 配对码画面（不保存） | 扫码配对 | 可选（可手动输入配对码时） | 不保存 |
| 照片 | 你选的照片（加密传输） | 发给 Agent | 可选 | 密文，取到即删 |
| 任务与记录 | 1.4 所列 | 同步与 Agent 上下文 | 使用该功能时必需 | 阿里云上海 |
| 简报 | 事件时间窗；引用的原话 | 回来时给出简报 | 可选 | 阿里云上海 |
| Agent Use（脚本路径） | 不涉及模型；每天经脚本成功的次数（不含内容） | 操作 Agent 应用 | Agent Use 开关，现有安装默认开（决定手机发话能否生效） | 用量见 4.3 |
| 模型辅助（Agent Use 的模型路径） | 控件列表、窗口截图（不存）、用量记录 | 脚本走不通时请模型决定下一步 | 可选（默认关闭，单独同意；各档位一样） | 不存；用量见 4.3 |
| 失败报告 | 1.6 所列字段 | 修复适配 | 可选（可关闭） | 阿里云上海，90 天 |
| 通知 | 推送令牌、时区 | 推送提醒 | 可选 | 中继与内容服务 |
| 发布链接 | 发布的文件、版本、同意记录、审核回执 | 发布公开页面 | 可选 | 阿里云上海 |
| 订阅 | 交易信息、档位、操作种类与时间计数 | 确定权益、计量 | 付费功能必需 | 阿里云上海 |
| 安全 | IP（运算值）、请求时间与结果 | 安全与防滥用 | 必需 | 阿里云上海 |
| 头像 | 一张小图 | 显示头像 | 可选 | 阿里云上海 |
附录二：合作方与第三方信息共享清单
知在 App 没有嵌入任何第三方 SDK，下列都是我们使用的服务。「受托」指受我们委托处理；「第三方」指按其自身规则处理。
| 合作方 | 做什么 | 涉及的个人信息 | 所在地 | 性质 |
|---|---|---|---|---|
| 阿里云（阿里云计算有限公司） | 服务器、数据库、对象存储、边缘加速（ESA） | 第一节所述服务端数据（附件与成果为密文） | 中国（上海）；ESA 节点按访问地 | 受托 |
| 阿里云短信服务 | 发送登录验证码 | 手机号码 | 中国 | 受托 |
| 阿里云内容安全 | 审核要发布的公开内容 | 发布内容中的文字与图片 | 中国（上海） | 受托 |
| 阿里云百炼（Agent Use 的模型） | Agent Use 慢路径的模型处理 | 控件列表、Agent 应用窗口截图（可能含窗口里显示的任何内容）；百炼按法律法规留存调用数据 | 中国 | 受托（另取得你对「模型辅助」的单独同意，见 1.5） |
| Apple（通过 Apple 登录） | 登录 | Apple 给出的用户标识 | 美国 | 第三方 |
| Apple（推送通知服务） | 通知、实时活动 | 推送令牌、1.7 所述推送内容 | 美国 | 第三方 |
| Apple（App Store） | 订阅与付款 | 交易信息 | 美国 | 第三方 |
附录三：你接入的 App 与 Agent
中国大陆区目前不开放远程连接的 App（见 1.10）。
你在 Mac 上运行的 Agent（见 3.4.2）
| Agent | 把记录交给 | 所在地 | 联系方式 |
|---|---|---|---|
| Claude Code | Anthropic PBC | 美国 | privacy@anthropic.com |
| Codex | OpenAI OpCo, LLC | 美国 | dsar@openai.com |
| Cursor | Anysphere, Inc.，并由它转给你选择的模型服务 | 美国 | hi@cursor.com |
附录四：权限使用说明
| 权限 | 用途 | 何时申请 | 不授权的影响 |
|---|---|---|---|
| 相机（iPhone） | 扫描 Awareful Desk 上的配对码 | 第一次扫码时 | 不能扫码配对 |
| 通知（iPhone） | 有事需要你时提醒 | 你开启提醒时 | 收不到提醒，其他功能不受影响 |
| Face ID（iPhone） | 批准需要你本人确认的操作 | 第一次需要时 | 不能批准这类操作 |
| 辅助功能（Mac） | 在 Agent 应用窗口里替你操作 | 你开启 Agent Use 开关或首次从手机发话时 | 不能替你在 Agent 应用里操作 |
| 屏幕录制（Mac） | 看清目标窗口、确认操作落实；离开时遮屏 | 同上 | 同上，且不能遮屏 |
English (Global region)
Effective date: November 4, 2026 Last updated: November 4, 2026
Who we are: Soudaku (Hainan) Technology Co., Ltd. (搜达库（海南）科技有限公司; "Awareful", "we", "us").
Postal address: 海南省海口市龙华区大同街道龙华路17号财盛大厦第16层C区C900号
Contact for privacy questions and requests: privacy@awareful.com
1. The short version
- Awareful (知在 in Chinese) helps you supervise coding agents such as Claude Code, Codex and Cursor that run on your Mac, from your iPhone. This policy covers the Awareful app on iPhone, Awareful Desk on your Mac, our relay and content services, our remote connector, the pages you publish, and awareful.com.
- Two layers of privacy. Your session mirror, the approvals and questions your agents raise natively, what you send to your agents, and your attachments and results are end-to-end encrypted between your Mac and your paired phone. Our servers see only ciphertext. Your tasks, records, ideas, questions your agents ask through Awareful, and messages are stored on our servers in a form we can read, so we can sync them across your devices and give your agents their working context.
- Agent Use sends screen images to an AI model, sometimes. When you have agreed to model assistance for Agent Use (Awareful Desk operating your agent apps for you; it is off until you agree, on every plan) and the scripted path cannot continue, Desk may send a list of on-screen controls, and if that is not enough, a screenshot of the agent app's windows (with any system dialog shown over them), to an AI model provider. Section 3.5 and section 6 explain who, why and what we keep. We do not store those screenshots. We turn on zero data retention at our model gateway for Global-region accounts.
- Records are not sent to a model for briefings. Your return briefing is computed by rules from events that happened. It is not written by an AI model.
- We do not sell your personal information, we do not share it for cross-context behavioral advertising, we have no third-party advertising or analytics in the app, and we do not track you across other companies' apps and websites.
- You can export your data and delete your account inside the app. Deletion takes effect immediately for sign-in and sync; we erase content after 7 days, and backups expire at most 7 days later.
- Awareful is not for children under 13.
2. Who this applies to
Accounts created from App Store storefronts outside mainland China are in the Global region. Your region is set when you create the account, from your App Store storefront, and does not change when you travel. Global-region data is stored in the United States (Amazon Web Services, Oregon, us-west-2). Accounts created from the mainland-China storefront are in the China mainland region, are stored in China and are covered by a separate Chinese-language policy.
Third-party products you choose to connect (for example ChatGPT, or the coding agents you run) are governed by their own terms and privacy policies. Section 5.3 and Appendix B explain where our responsibility ends.
3. Information we collect, and why
3.1 Account and sign-in
| What | Details |
|---|---|
| Sign-in identity | Sign in with Apple, Google, GitHub or an email code. We receive a signed identity token and keep only the provider's user identifier. We do not store your name or email from Apple, Google or GitHub, and we do not keep the provider's access tokens. |
| Email code | We use your email address once, to ask Amazon Simple Email Service (US) to send the code. We do not keep the address in clear; we keep a keyed hash (to recognise the same address) and a masked label (for example l***@gmail.com). Codes expire after 5 minutes. |
| Region | The three-letter App Store storefront code, to decide your region. |
| Devices | Device model name (for example "iPhone 17 Pro"), a device identifier and pairing records. We do not store the IP address you connect from; it is used for rate limits, and for sign-in abuse checks we keep only a keyed hash (below). |
| Abuse prevention | Keyed hashes of the address, IP and an app-generated random device id, plus send time, kept 24 hours. No raw address or IP is kept for this. |
You can link several sign-in methods to one account and unlink them; you must keep at least one. Without one of them you cannot create or use an account.
3.2 Pairing, session mirror, approvals and sending (end-to-end encrypted)
- Pairing. The phone scans a pairing code on Awareful Desk (the camera is used only for scanning). Phone and Mac exchange public keys. Our relay records device identifiers, public keys, the pairing, the push token and when each device was last seen. It holds the phone's name only while a pairing waits for you to confirm it on the Mac, and it does not store IP addresses. The first time a new phone is paired you confirm four digits on the Mac.
- Session mirror. Awareful Desk reads the Claude Code, Codex and Cursor sessions in the scope you choose on your Mac, encrypts them on your Mac, and sends them through our relay to your paired phone. Only your Mac and your paired phone can decrypt them.
- Approvals and questions. What an agent natively asks you (a command to approve, a question) is delivered encrypted. Our servers keep only a request id, the task and source session it belongs to, whether it can only be answered on the Mac, expiry, and status (when it was answered and delivered), not the command, question or answer text. Your answer is signed with a key that stays on your iPhone; actions that need you to confirm in person use Face ID.
- Sending and attachments. Messages, photos and files you send to an agent, and files, images, PDFs and web pages an agent hands to your phone, travel encrypted. We store ciphertext until every receiving device has fetched it. Photos come through the system photo picker; we receive only the ones you pick.
3.3 Tasks, records and context (readable by our servers)
This is the core of the product. The following is stored on our servers and we can read it:
- tasks and subtasks (titles, your own words, what "done" means, state, reminders, links between tasks), including to-dos that are just for you;
- record entries that you or an agent write (decisions, changes of direction, results, risks, handoffs) with their revision history;
- ideas and bookmarks (the text you write; any session excerpt attached to a bookmark is encrypted on your device and we keep only ciphertext);
- requests: the questions agents ask you through Awareful's own tools and your answers, in clear text;
- messages you leave for a session or an area, and when a session picked them up;
- preferences, skills you save, agent suggestions and feedback; areas and tidy-up suggestions and your choices;
- provenance: which agent (Claude Code, Codex, Cursor, ChatGPT and so on) wrote an entry, in which session (session identifier and title);
- your actions and focus: the last thing you did on a task and when (with your own words or a short name), the task and session you are focused on, when you left and came back;
- modes (Build, Wild, Sleep): current mode, switch times and source, wake-up time, time zone, reminder settings;
- the folder paths of the working trees and git repositories behind your sessions (paths only, never code), reported by Awareful Desk to organise suggestions;
- an optional avatar (one small image, 128 KB at most, with location metadata removed);
- reading positions: when you last finished reading your briefing, and how far you have read the activity feed in each scope (all, one task, one agent on one Mac), up to 200 scopes; ids and times only, no content;
- a log of agent and app calls: for each call an agent or connected app makes, the tool, the shape of its arguments and any search words, how many results came back, the ids it touched, the verified session, timing and outcome; never entry bodies. Kept 30 days; you can see it in the app under Me.
Briefings and quoted lines. When you come back (from Wild or Sleep) we freeze a briefing (called a sitrep in the app) from events that really happened. It can contain your own words about a task, ideas you noted, questions and answers, messages, and short lines quoted from the mirrored session, such as the last thing you said before you left or something you said on the scene. We store these quoted lines on our server in readable form and make them searchable for you. We do not store whole sessions this way, and native approval and question contents and answers are not included. If you undo a switch back, or delete the task or idea a quotation belongs to, the corresponding text is cleared. The briefing is composed by rules, not by an AI model.
What you and your agents write is up to you. If a record contains sensitive personal information about you or other people (health, finances, location, credentials), we store and sync it like any other text and it may be handed to connected agents and apps as described in section 5. Please do not put passwords or keys in records, and only include other people's personal information if you are entitled to.
3.4 Awareful Desk on your Mac
Awareful Desk needs macOS Accessibility and Screen Recording permission to carry out your instructions inside the agent apps: send a message, create or fork a session, change session settings, answer a native question group. Desk prefers the app's accessibility controls and uses targeted background input only for that window; it does not move your pointer. It also uses screen recording to cover your screen while you are away (this is not a lock screen).
- Run logs (stages, outcomes, diagnostic events) and the images Desk sees while acting stay on your Mac. They are not uploaded, except what Agent Use sends as described in 3.5 and 3.6.
- The receipts sent to your phone contain stages and outcomes only, not message text, and are end-to-end encrypted.
- Desk reports to us whether the Mac is covered or locked, when you left and came back, the working-tree paths mentioned in 3.3, and the metadata of native approvals and questions described in 3.2.
- The descriptions of agent-app controls that Desk relies on ship inside Desk; Desk does not download them from us.
- You can revoke the permissions at any time in macOS System Settings; Desk then cannot act or cover the screen.
3.5 Agent Use: scripted path and model path
Agent Use from your phone comes with the Plus and Pro plans, including a new account's first 14 days of Plus. The data handling below applies to every account, Free included whenever it uses the model path; where this policy says "model assistance" it means the model path for any account. Agent Use has three layers, in this order: the agent app's own channel, then verified scripts, then an AI model.
Scripted path (most operations). Everything stays on your Mac. No model is involved.
Model path ("slow path"). Only when a script cannot continue, Desk may ask a model what to do next. This happens in two steps:
- Desk sends a list of controls visible in the target app window: each control's role, name and position. Session titles and draft text in that list are replaced with placeholders.
- Only if the list is not enough, Desk sends a screenshot of the target agent app's windows on that display, including any system dialog shown over them (for example a file picker or a permission prompt). Other apps' regular windows and Awareful's own windows are left out. A screenshot can show anything those windows and dialogs show: your conversation, code, file names, folder paths, other apps' prompts, personal information, secrets. Awareful Desk does not choose what is visible; please do not use the model path while a window shows something you do not want an AI provider to process.
How it travels (Global region): your Mac -> our content service (AWS, US) -> OpenRouter, Inc. (US) -> an AI model provider. We set OpenRouter's zero-data-retention requirement (provider.zdr: true) on every request, so requests are routed only to endpoints that do not store prompts or outputs. Under OpenRouter's published terms, ZDR means the provider will not store your data for any period; OpenRouter itself says it does not retain prompts or responses for such requests, with narrow exceptions for legal and child-safety compliance.
What we keep: we do not write screenshots or control lists to disk or database. We keep a record of each model call to enforce usage caps: the run it belongs to, the host app, the step, the model, token counts, cost, duration, how many controls were listed and the screenshot's size in bytes; never the controls, the screenshot, the task or the model's answer. We delete these records after 90 days, and with your account.
Safeguards: the model path only performs reversible actions (find a session, scroll, close a menu, focus the input, type the text, which can be cleared before sending). Before any send that cannot be undone, Desk checks deterministically that it is in the right session and that the full text and attachments are correct. A task is reported done only on the host app's own receipt.
Consent. There are two separate controls. (1) The Agent Use switch (Desk, and the phone's Mac settings) lets the phone's sends, new sessions and settings changes act in your agent apps through the scripted path; it does not send anything to an AI model. (2) Model assistance is a separate choice that is off until you agree on its own screen, which names the recipients (our gateway, OpenRouter and the model provider behind it) and says that a screenshot can show anything in the window. Without (2), Desk uses only the app's own channel and verified scripts, and hands the task back to you when they cannot continue. We record which version of that screen's text you agreed to and when, and our service refuses any model call for an account that has not agreed to the current version. You can withdraw (2) at any time, next to the Agent Use switch.
Automated decisions. The model chooses clicks in an app on your own computer. It does not make decisions about you with legal or similarly significant effects.
3.6 Failure reports (Agent Use)
When reports are on for a Mac and something goes wrong or the script hands over to the model (script gives up, a pre-send check fails, the model path ends, an outcome is "unverified" or "needs a human"), Desk sends us a small report so we can fix our adapters. A script-only success is not reported individually; Desk sends one daily count.
Fields: run id and time; host (codex, claude-code, cursor), the host app's version and build, its UI language; Desk, kernel and adapter versions; kind of operation; route taken (the app's own channel, script, model); why the script handed over (a code such as "no composer found") and at which step; model usage (number of decisions and screenshots, cost, provider, model); which pre-send check failed; the outcome and its reason code; and, for each control our adapter expects, whether it was found, missing or ambiguous.
Not included: message text, titles, session ids, screenshots, raw accessibility labels, model reasoning, file names.
We keep reports for 90 days, use them only to find and fix adapter breakage (our operators read counts per app version), and delete them when you delete your account. Reports start off on a Mac whose region in macOS (System Settings > General > Language & Region) is in the European Economic Area or the United Kingdom, and on everywhere else. You can turn them on or off for each Mac: in Awareful Desk under Privacy (Failure reports, below the Agent Use switch), or on your phone in that Mac's card (Turn off reports, or Turn on reports while they are off). Your choice replaces the default. While off, that Mac sends no reports, and turning them off drops any it has not sent yet.
3.7 Notifications, Live Activities and widgets
If you allow notifications we store your push token and send pushes through Apple's push service. A push carries only identifiers, fixed text and numbers (for example a request id and "Something needs you"), never session content, questions, answers or task names. Live Activity updates for the Lock Screen and Dynamic Island that go through Apple carry only counts (needs you, in progress, done today, needs attention), Desk's state, when the screen cover started, whether a Mac is paired, and the kind and time of the latest step (done, stuck, asked, step); never titles or text. On your phone, the app then adds the task's short name, the session title and what the request is (for example "Needs you 1 · Prototype C · Claude wants to run a command"), which anyone who can see your Lock Screen can read. Home Screen widgets read data the app keeps on your phone. Your phone reports its time zone so reminders arrive at local time.
3.8 Publishing a link
You can publish a result as a public link. An agent can publish only with your own words from that conversation or after you tap Allow on your phone. We process the files (stored in clear in a separate bucket in your region), title, versions, source session and task, and the record of your consent. Anyone with the link can see it; the random address is not privacy. After takedown the link stops working at once, but we cannot recall what others saw or saved. Global-region uploads get a static anti-phishing check; there is no pre-review. Public pages set no cookies and we do not log visitors' IP addresses or identities; for each account we count the bytes and requests its links serve each month, for the traffic limit, with no visitor data.
3.9 Plans, purchases and usage
We record the kind and time of each action you take on your phone (approve, answer, send, new session), and your plan at the time, never the content, to decide entitlements and prevent abuse. For a new account's first 14 days of Plus, from pairing a Mac, we record when they start and end. Subscriptions are processed by Apple; we never see your card or billing address. We receive Apple-signed transaction data (product, transaction ids, environment, storefront, status, expiry) to decide your entitlements. Apple sends its purchase notices for the whole app to one address, our China mainland server, which passes notices for non-China storefronts on to the Global region without keeping them. For purchases on a non-China App Store storefront, RevenueCat processes your purchase records (the transactions above, keyed to your account id, not your name or email) for us, for revenue reporting; it does not decide what you can use. We keep a few fields of RevenueCat's purchase events (type, product, environment, times, price in US dollars) for up to two years, deleted with your account. Purchases on the China storefront never reach RevenueCat.
3.10 Connected apps
When you connect Awareful inside another app (for example ChatGPT) we process the app's registration, the authorization request, check digits you enter on your phone, one essential cookie in your browser (5 minutes), your separate consent record (text version, time, which phone), and the tokens we issue to that app (stored only as digests; access tokens last at most one hour; a connection lapses after 90 days unused).
3.11 Security, diagnostics and support
We process IP addresses (a keyed hash where used for rate limits), request times and results to keep the service secure. Server logs never contain record text or keys. Servers in the Global region delete them after 30 days; servers in the China mainland region keep them 190 days, as Chinese law requires network logs to be kept at least six months. iOS crash and hang reports stay on your phone; we receive a diagnostics package only if you export and send it, and it contains no private keys, tokens, session text or anything you wrote. If you email us, we process your address and what you tell us.
3.12 On your devices only
Your phone keeps a local full-text search index of mirrored conversations (what you and agents said; not tool calls or output) so you can search; it never leaves the phone and is deleted with the mirror. Answered native questions and your "likes" on activity items stay on the phone. Caches of results and attachments are capped at 500 MB and can be cleared. Your devices also keep where you last read up to and sync it between your devices through your own iCloud key-value storage, which we cannot read; our server keeps reading positions too (3.3).
3.13 Demo mode
The app includes a demo mode, so reviewers and you can look around and show friends without an account or a Mac. Demo mode uses fictional sample data bundled in the app, kept in its own storage apart from your real account, which it cannot read or change. The app answers its own requests from that sample data on your phone and refuses every other request, so nothing goes over the network and nothing you do in demo mode reaches us. It also leaves iCloud, purchases and notifications alone.
3.14 What we do not collect
Precise location, contacts, advertising identifiers, browsing history, photos you did not pick. We use no third-party analytics, advertising or attribution SDK; the iPhone app embeds no third-party SDK. Apps and websites of ours do not track you across other companies' apps and websites.
Website, Do Not Track and Global Privacy Control. awareful.com is a static site with no cookies and no analytics or advertising tools. It is hosted on our servers in China and delivered through Alibaba Cloud's ESA edge network, which has edge servers worldwide, including in China; a visitor's IP address and request metadata pass through those servers to deliver the page. Our web server keeps no access log of site visits, and we have not turned on the edge network's log delivery, so we receive no visitor logs from it; Alibaba Cloud processes that request data under its own terms. Because we do not track you or sell or share personal information, there is nothing for a Do Not Track signal to switch off; we honour a Global Privacy Control signal as an opt-out of sale and sharing, although we do neither.
4. How we use information, and our legal bases (EEA/UK)
| Purpose | Legal basis (GDPR / UK GDPR) |
|---|---|
| Create and run your account, sync, deliver approvals and messages, run briefings, publish links | Performance of a contract (Art. 6(1)(b)) |
| Agent Use model path | Your consent to model assistance, given on its own screen (Art. 6(1)(a)); contract for the scripted path |
| Push notifications, camera, Face ID, connected apps | Consent, given through the system prompt or our separate consent screen; withdraw any time |
| Security, abuse prevention, rate limiting, server logs | Legitimate interests (Art. 6(1)(f)) in a secure service |
| Failure reports | Your consent when you turn them on (they start off on Macs set to an EEA or UK region); legitimate interests in keeping Agent Use working elsewhere |
| Plan, entitlement and tax/accounting records | Contract; legal obligation |
| Responding to you and to legal requests | Legitimate interests; legal obligation |
Where processing relies on consent you can withdraw it at any time without affecting earlier processing. Providing sign-in data is necessary for an account; the rest of the information is needed only for the feature that uses it. We do not use your content to train AI models. On the Agent Use model path we route requests only to endpoints that keep no data (3.5).
5. Who we share information with
We do not sell personal information and do not share it for cross-context behavioral advertising (as those terms are used in US state privacy laws).
5.1 Service providers
They process data on our behalf under contract (full list in Appendix A): Amazon Web Services (hosting, database, storage, CloudFront for published links, SES for email codes); Alibaba Cloud ESA (edge entry point mcp.awareful.com for remote connections; it forwards each request to your region and we do not process or store your content in another region); OpenRouter, Inc. and the model providers it routes to (Agent Use model path, section 6); RevenueCat, Inc. (purchase records for non-China storefronts, section 3.9).
5.2 Platforms you use
Apple (Sign in with Apple, push notifications, App Store purchases), Google and GitHub (if you sign in with them). They act under their own policies.
5.3 Agents and apps you connect
- Remote connectors (for example ChatGPT or Claude). Before the first connection, the confirmation screen in the Awareful app tells you who receives your data, where, what, why and how to withdraw, and asks for your separate consent. Within the scope you grant (read, or read and write) the app can read your tasks and records (which can contain personal information about you and others). With write access it can also add records labelled as coming from it, change entries AIs recorded, replace or withdraw your decisions with a reason that you see and can undo, manage the steps of tasks, and send answers to your phone as notifications. It cannot rewrite your own words, operate your Mac, approve requests for you, change settings or export your account. Disconnect under Me > Account > Connected apps. Disconnecting does not recall what it already received.
- Agents you run on your Mac (Claude Code, Codex, Cursor and others). Awareful Desk installs a connector so these agents read and write your tasks and records. They are third-party software that you run with your own accounts. What they read goes, with the rest of your conversation, to their own model providers (Anthropic, OpenAI, Anysphere and the model you pick in Cursor), mostly in the United States, under their policies (Appendix B). You can remove an agent in Desk under Connect agents.
- Their handling of data after they receive it is theirs, not ours.
5.4 Legal, safety and business transfers
We may disclose information to comply with law or valid legal process, to protect rights, safety and the service, or in a merger, acquisition or asset sale (with notice, and the recipient must honour this policy).
6. AI models and third-party AI: everything in one place
This section exists so you can see every place where information leaves your control and goes to an AI model.
- Agent Use model path (ours). Control lists, and when needed a screenshot of an agent app's windows, go through our gateway to OpenRouter and a zero-retention model provider (3.5). It applies to any account that uses the model path; off until you agree to model assistance on its own screen, on every plan, and you can withdraw at any time.
- Agents you run. What they read from Awareful goes to their model providers (5.3).
- Connected apps (ChatGPT). What you authorise goes to that app and its provider (5.3).
- Not sent to a model: briefings (computed by rules), the session mirror, native approvals and answers, attachments and results (end-to-end encrypted).
We do not use your information to train AI models. Our provider list may change; we update Appendix A and tell you in the app before any change that sends new categories of data to a new kind of recipient.
7. International transfers
We store Global-region data in the United States. Our operator is a company in China, and the people who run the Service may access Global-region data from China, only as far as needed to operate, secure and support it. Where the law requires, we rely on standard contractual clauses or another valid mechanism; you can ask for a copy through the contact details above. When your Mac or phone is outside the United States your data still travels to your region's servers.
8. How long we keep information
| Information | Kept |
|---|---|
| Tasks, record entries, ideas, bookmarks, areas | Until you delete them; then 30 days in Recently Deleted, then erased with revisions |
| To-dos and subtasks | Until you delete them; deleting is immediate and permanent |
| Requests, messages, handoffs, preferences, skills, agent feedback, your actions and focus, modes and away records, briefings (including quoted lines), reading positions, avatar, action counts, monthly traffic counts for published links | Life of the account, deleted with it; unclaimed area messages expire after 7 days |
| Tidy-up suggestions and your choices | Until account deletion; adoptions can be undone for 30 days |
| Encrypted session mirror, approvals, answers (relay) | Deleted once every receiving device confirms; undelivered mirrors at most 30 days; approvals and answers at most 1 day |
| Encrypted attachments and results (relay or object storage) | Deleted once every receiving device fetches them; at most 7 days; unfinished uploads removed after 24 hours |
| Attachments from your phone, on your Mac | 7 days; deleted at once if the phone is unpaired |
| Results on your Mac | Latest version until you delete it; earlier versions 90 days |
| Finished approvals and sends on your phone | 7 days |
| Published links | Current version until you take it down; other versions 90 days; restorable 7 days after takedown, then site, versions and files are deleted |
| Agent and app activity log (MCP calls) | 30 days |
| Write-retry receipts | 30 days |
| Server logs (no record text or keys) | 30 days in the Global region (190 days in the China mainland region) |
| Agent Use failure reports | 90 days |
| Agent Use screenshots and control lists | Not stored by us (they exist only while a request is processed; see 3.5 for the provider's side) |
| Agent Use model-call records | 90 days; deleted with your account |
| Agent Use consent record (agree or withdraw, text version, time, phone or Desk) | Life of the account; deleted with it |
| Daily counts of scripted Agent Use successes per Mac (host, app version, count; no content) | 90 days |
| First 14 days of Plus record (start and end) | Life of the account; kept after the 14 days end; deleted with the account |
| Sign-in identities | Deleted when you unlink; with the account otherwise |
| One-time codes and authorization flows | Email codes 5 minutes; sign-in flows (Apple, Google, GitHub) 10 minutes; connected-app authorization requests 5 minutes; cleaned up within an hour; abuse counters 24 hours |
| Push tokens | When a device is removed, you sign out, the account is deleted, or Apple reports the token invalid |
| Connected-app tokens | Access tokens 1 hour; lapse after 90 days unused; revoked on disconnect or deletion |
| Database backups | Rolling 7 days; deleted data is gone from backups within 7 days after deletion |
Data on your Mac and phone is removed when you uninstall. After the periods above we delete or anonymise information, unless the law requires us to keep it.
9. Security
TLS on all connections; end-to-end encryption for the content listed in 3.2; signing keys that never leave your iPhone; Face ID for actions that need you in person; hashes instead of raw phone numbers, emails and IPs; token digests; logs without content or keys; regions isolated from each other; published pages sandboxed from account data. No system is perfectly secure. If a breach affects you we will notify you and regulators as the law requires.
10. Your choices and rights
In the app, without contacting us: view and export your data (Me > Account > Export, a JSON file); edit or delete records, ideas, bookmarks, to-dos and your avatar; take down published links; remove a Mac or phone; disconnect connected apps; remove agents in Desk; turn Agent Use off and withdraw model assistance; revoke notification, camera, Face ID, Accessibility and Screen Recording permissions in system settings; delete your account (Me > Account > Delete account). Deletion takes effect at once for sign-in, pairing, relay content and links; signing back in with the same method within 7 days cancels it; after 7 days we erase the account and its content, and backups expire at most 7 days after that (14 days at the outside). If you have a paid subscription, deleting the account does not cancel it: cancel it in Apple's subscription settings first.
If you are in the EEA, UK or Switzerland: you have the rights of access, rectification, erasure, restriction, portability and objection, to withdraw consent, and to lodge a complaint with your supervisory authority. If you live in California or another US state with a privacy law: you have the rights to know, access, delete, correct, and opt out of sale or sharing, to limit use of sensitive information, and to not be discriminated against for using these rights. We do not sell or share personal information and do not use sensitive personal information to infer characteristics, so there is nothing to opt out of or limit beyond what the product settings give you. Contact us at the address above; we will verify your identity (usually by asking you to start from your signed-in account), reply within 45 days (30 days in the EEA/UK, extendable as the law allows), and tell you why if we decline. You may use an authorised agent. If we refuse, you may appeal by replying to our message. California residents: see Appendix C. Nevada residents: we do not sell covered information.
11. Children
Awareful is not directed to children under 13 and we do not knowingly collect personal information from them. If the law where you live sets a higher age for consenting to the processing of your personal information (up to 16 in parts of the EEA), you need a parent or guardian's agreement below that age. If we learn that a child under 13 has given us personal information we will delete it. Contact us if you believe this has happened.
12. China mainland accounts
If your App Store storefront is mainland China, your account is in the China region and the Chinese-language policy in the first part of this page (awareful.com/privacy) applies, not this one.
13. Changes
If we make material changes (new categories of data, new kinds of recipients, new purposes) we will tell you in the app before they take effect and update the date above. Earlier versions are available on request.
14. Contact
Soudaku (Hainan) Technology Co., Ltd., at the postal address at the top of this policy; privacy@awareful.com. We have not appointed a representative in the EU or the UK; write to privacy@awareful.com from anywhere. Support: support@awareful.com.
Appendix A. Service providers
| Provider | What for | Personal information involved | Location |
|---|---|---|---|
| Amazon Web Services | Servers, database, S3, CloudFront (published links), SES (email codes) | Server-side data in section 3 (attachments and results as ciphertext); email address for sending a code | United States (Oregon); CloudFront edge by visitor location |
| Alibaba Cloud ESA (overseas edge) | Entry point mcp.awareful.com for remote connections (forwards to your region), and edge delivery of the website www.awareful.com | Request metadata and authorization data in transit; website visitors' IP addresses and request metadata | Edge nodes by visitor location |
| OpenRouter, Inc. | Gateway for the Agent Use model path | Control lists, screenshots (in transit; zero data retention requested) | United States |
| Model providers behind OpenRouter | Run the model for each Agent Use request | Same | Per provider, chosen by OpenRouter for each request among zero-data-retention endpoints |
| RevenueCat, Inc. | Purchase records and revenue reporting, non-China App Store storefronts only (3.9) | App Store transactions and Apple's purchase notices, keyed to your account id | United States |
| Apple | Sign in with Apple, push notifications, App Store purchases | Identity token; push token and the minimal push content in 3.7; transactions | United States and worldwide |
| Google, GitHub | Sign-in, if you choose it | Provider user identifier | United States |
Appendix B. Apps and agents you connect
| Product | Operator | What it receives | Contact |
|---|---|---|---|
| ChatGPT (connector, with your separate consent) | OpenAI OpCo, LLC, United States | The records you authorise (read, or read and write) | privacy.openai.com; dsar@openai.com |
| Claude (connector, with your separate consent) | Anthropic PBC, United States | The records you authorise (read, or read and write) | privacy@anthropic.com |
| Claude Code (agent you run) | Anthropic PBC, United States | What it reads from Awareful, with your conversation | privacy@anthropic.com |
| Codex (agent you run) | OpenAI OpCo, LLC, United States | Same | dsar@openai.com |
| Cursor (agent you run) | Anysphere, Inc., United States, and the model provider you pick | Same | hi@cursor.com |
Appendix C. California notice at collection
We collected the following categories in the past 12 months. We did not sell or share any of them, and we do not use or disclose sensitive personal information for purposes beyond those allowed by the CCPA regulations.
| CCPA category | What | Source | Purpose | Disclosed to |
|---|---|---|---|---|
| Identifiers | Provider user id, device id, device model name, IP (hashed where stored), masked email label, push token | You, your devices | Account, security, notifications | Service providers (Appendix A), Apple |
| Commercial information | Apple transaction data, plan, action counts | Apple, your use | Entitlements, usage limits | AWS |
| Internet or other electronic network activity | Agent/app activity log, action counts, server logs | Your use | Service, security | AWS |
| Audio, electronic, visual information | Avatar; photos and files you send (encrypted); Agent Use screenshots (not stored by us) | You, your Mac | Features you use | AWS (ciphertext); OpenRouter and model providers (screenshots, in transit) |
| Professional or content-related information | Tasks, records, ideas, questions and answers, quoted briefing lines | You, your agents | Core service | AWS; connected apps and agents you authorise |
| Inferences | None | n/a | n/a | n/a |
Retention: section 8. We do not knowingly sell or share personal information of consumers under 16.