先获得授权，
再理解上下文。Permission first.
Context second.

宇瀚（深圳）人工智能有限公司（宇瀚人工智能）负责本网站和 AI替身私测服务的数据处理。Yuhan (Shenzhen) Artificial Intelligence Co., Ltd. (Yuhan AI) is responsible for data processing for this website and the AI替身 private alpha.

ceo@yuhandigital.com

不同客户端和授权状态会产生不同数据。可能包括：The data varies by client and permission state. It may include:

• 账号、设备、会话与安全审计信息；account, device, session and security-audit information;
• 用户明确提交的目标、原则、笔记、反馈和结果；goals, principles, notes, feedback and outcomes submitted by the user;
• 经单独授权的应用使用元数据、通知、日历、浏览活动、文件或剪贴板片段；separately authorized app-usage metadata, notifications, calendars, browsing activity, files or clipboard excerpts;
• 当前可见界面文字、AI替身输入组件中的输入、用户主动分享的内容；currently visible interface text, text entered through AI替身 input components, and content the user actively shares;
• 用户启动相应功能后产生的屏幕文字识别或语音转写；screen-text recognition or voice transcripts produced after the user starts those features;
• 为诊断可靠性、安全性与滥用而生成的技术日志。technical logs used to diagnose reliability, security and abuse.

AI替身不能读取密码框、安全窗口、未显示的应用内容或其他应用的私有数据库，也不能保证取得微信等应用的完整历史。AI替身 cannot read password fields, secure windows, content that has not been displayed, or another app’s private database. It cannot promise complete history from apps such as WeChat.

• 把新信息与用户确认的目标联系起来；connect new information to goals the user confirmed;
• 发现可能的风险、偏差、机会或待办；identify possible risk, deviation, opportunity or unfinished work;
• 生成、投递、去重和校准建言；generate, deliver, deduplicate and calibrate counsel;
• 在用户设备间同步目标、建言与反馈；synchronize goals, counsel and feedback across the user’s devices;
• 防止越权、滥用、重复提醒和模型费用失控；prevent unauthorized access, abuse, repeated notifications and uncontrolled model spending;
• 改进私测产品的准确性、时机与可靠性。improve accuracy, timing and reliability during the private alpha.

当你预约产品演示、索取投资资料或申请参与私测时，我们会处理你主动填写的姓名、邮箱、公司或机构、职位、申请类型、场景说明，以及你选择提供的平台偏好或个人主页链接。我们还会保存同意记录，并为防止垃圾提交和保障服务可靠性生成最少量的技术哈希；网站申请数据库不保存原始 IP 地址。When you request a demo, ask for investor materials or apply for the private alpha, we process the name, email, organization, role, request type and scenario details you choose to provide, together with any platform preference or profile URL. We also retain a consent record and generate minimal technical hashes to prevent spam and maintain reliability; the website inquiry database does not store raw IP addresses.

这些信息只用于回复申请、评估演示或合作适配度、安排后续沟通，以及保护表单免受滥用。公开网站、表单接口和申请记录由 Cloudflare Pages、Workers 与 D1 提供支持。为及时通知并回复申请，完整表单内容会通过 Google Apps Script 与 MailApp 转发到固定的公司邮箱 ceo@yuhandigital.com；该邮箱由 Zoho 承载。Cloudflare、Google 和 Zoho 的处理可能发生在你所在国家或地区以外。我们会定期检查申请记录，并在不再需要时删除；你也可以通过该邮箱申请访问、更正或删除。We use this information only to respond, assess fit for a demo, partnership or alpha, arrange follow-up, and protect the form from abuse. The public site, form endpoint and inquiry records are supported by Cloudflare Pages, Workers and D1. To notify our team and reply promptly, the complete form submission is forwarded through Google Apps Script and MailApp to the fixed company inbox ceo@yuhandigital.com, which is hosted by Zoho. Processing by Cloudflare, Google and Zoho may occur outside your country or region. We periodically review inquiry records and delete them when they are no longer needed; you may also request access, correction or deletion at that address.

主动云分析默认关闭，需由用户开启。开启后，完成判断所需的内容会先进入宇瀚人工智能控制的日本区域私测服务器，并可能经第三方 OpenAI 兼容 API 网关发送给所选上游模型服务商。网关和上游服务商都可能接收推理所需的文本与请求元数据；这不等同于内容一定直接发送给 OpenAI。Proactive cloud analysis is off by default and must be enabled by the user. When enabled, content required for a judgment first reaches Yuhan AI’s private-alpha server in Japan and may then pass through a third-party OpenAI-compatible API gateway to the selected upstream model provider. Both the gateway and upstream provider may receive inference text and request metadata; this does not mean the content is necessarily sent directly to OpenAI.

私测服务器、运营团队和模型处理方可能位于不同国家或地区，因此云分析可能构成跨境传输。当前处理角色与范围维护在The alpha server, operating team and model processors may be located in different countries or regions, so cloud analysis may involve cross-border transfers. Current processing roles and scope are maintained on our 处理方与数据流页面Processors and Data Flow page；用户应在开启云分析或完整原文上传前查阅。; users should review it before enabling cloud analysis or full-text upload.

某些私测版本允许用户进一步选择发送完整原文。该选择与普通云分析分开，关闭后不应继续发送新的完整原文。即使开启完整原文，密码、令牌、银行卡和常见凭据仍会经过出站净化规则。Some alpha builds let users separately choose to send full original text. This choice is distinct from ordinary cloud analysis, and disabling it should stop new full-text uploads. Even when full text is enabled, outbound filters still target passwords, tokens, bank-card details and common credentials.

我们使用传输加密、按账号隔离、逐设备会话、访问控制、备份和安全审计来降低风险。Windows 和 Android 客户端分别使用操作系统提供的安全能力保护本地会话凭据。We use encrypted transport, account-level isolation, per-device sessions, access control, backups and security audits to reduce risk. Windows and Android clients use operating-system security facilities to protect local session credentials.

账号、目标、事件、建言、反馈和转写文本通常在账号存续期间保存，直至用户删除账号或提出经核验的删除请求。发送到本地语音转写接口的原始音频只用于本次转写，不写入服务端业务数据库；生成的转写文本若被同步，则按事件记录保存。账号导出临时文件会在传输完成、超时或服务重启清理时删除。Accounts, goals, events, counsel, feedback and transcript text are generally retained while an account remains active, until the user deletes the account or makes a verified deletion request. Raw audio sent to the local speech-transcription endpoint is used for that transcription and is not written to the server’s application database; a synchronized transcript is retained as an event. Temporary account-export files are deleted after transfer, timeout, or startup cleanup.

私测系统仍可能出现缺陷，任何在线系统都无法承诺绝对安全。线上账号删除会清除该账号的活动租户数据；现有访问受控备份目前没有对外承诺固定的自动到期周期，残留副本可能保留到相应备份被安全替换或销毁。第三方模型处理方对请求的保存期限受其当时适用的条款约束。The private alpha may still contain defects, and no online system can promise absolute security. Account deletion removes that account’s active tenant data. Existing access-controlled backups do not yet carry a promised fixed automatic expiry, so residual copies may remain until the relevant backup is securely replaced or destroyed. Retention by third-party model processors is governed by their then-applicable terms.

• 逐项开启或关闭采集、云分析和完整上下文；enable or disable collection, cloud analysis and full context separately;
• 暂停通知、退出账号或撤销单台设备会话；pause notifications, sign out, or revoke a single device session;
• 查看并纠正目标、建言和反馈；view and correct goals, counsel and feedback;
• 申请导出或删除账号数据。request an export or deletion of account data.

如客户端暂未提供某项界面，请发送邮件至 ceo@yuhandigital.com。我们会先核验账号归属，通常在 30 天内回应访问、导出、更正或删除请求；如因复杂性或法律义务需要更长时间，我们会说明原因。If a control is not yet exposed in a particular client, email ceo@yuhandigital.com. We first verify account ownership and normally respond to access, export, correction or deletion requests within 30 days; if complexity or legal obligations require longer, we will explain why.

设备内容可能包含同事、家人、客户或聊天对方的信息。用户只应在拥有合法权利、获得必要同意且符合当地法律时启用相关采集或上传。AI替身不会因为设备属于用户，就假定用户能代表其他人授权。Device content may contain information about colleagues, family members, customers or conversation partners. Users should enable collection or upload only when they have a lawful basis, obtain necessary consent and comply with local law. Device ownership does not automatically authorize a user to act for other people.

AI替身私测面向成年人，不面向未满 18 周岁的儿童。我们会随产品和法律要求更新本政策，并在本页标示新的生效日期；重大变化会通过合理方式通知私测用户。The AI替身 private alpha is intended for adults and not for children under 18. We may update this policy as the product and legal requirements change, with a revised effective date on this page. Material changes will be communicated to alpha users through reasonable means.