Last updated: 2026-09-22
The extension sends the text of posts you are looking at to our server to be classified,
does not keep the text, collects nothing that identifies you, and sells nothing.
While you browse a supported site (x.com / twitter.com, xiaohongshu.com, weibo.com), the extension
takes the following from each post that enters your viewport and sends it to our server:
| Field | Notes |
|---|---|
| The post's text | Truncated to 1,200 characters |
| The author's public display name | Helps tell an official brand account from a person |
| Hashtags | Publicly visible |
| Whether the post links off-site | A true/false flag |
| Platform name | twitter / xiaohongshu / weibo |
Our server forwards that to the Jev model from TypeSafe AI, gets a verdict back and returns it to
the extension.
Clicking a label to report a mistake saves: the verdict, the label you say is correct, the
probability values behind that call, your sensitivity setting and the platform name.
The report panel has a checkbox, on by default, reading "Include this post's text". With it
ticked the post's text (truncated to 600 characters) and the author's public display name are saved
too, so the call can be reviewed. Unticked, only the numbers above are saved and no post content
is kept.
Reporting is entirely voluntary — nothing is sent unless you click. Reports are kept for one year.
no post text and no user identifier — when a popular post reaches many readers it is judged once.
to enforce the device limit. That identifier is generated by the extension and is tied to nothing
about you, your account or your hardware.
Post text is dropped from memory once the verdict is returned. It is never written to storage and
never written to a log.
Classification is performed by the Jev model from TypeSafe AI. Per their
published documentation, TypeSafe does not train on customer requests or responses. See their
Payments are handled by Creem, which acts as merchant of record. We never see your card details.
| Permission | Why |
|---|---|
storage | Keeps your settings and the local verdict cache |
| Host access to the three sites | Lets the content script read post text and place labels |
| Host access to our API domain | Lets the extension send text to be classified |
The extension does not request tabs, history, cookies or webRequest, and does not use
<all_urls>.
Email: jevadguide@gmail.com
最后更新：2026-09-22
本扩展把你正在浏览的帖子文本发送到我们的服务器做广告判定，不保存原文，
不收集任何能识别你身份的信息，不出售任何数据。
当你在受支持的网站（x.com / twitter.com、xiaohongshu.com、weibo.com）浏览时，
扩展会对进入可视区域的帖子提取以下内容，发送到我们的服务器：
| 字段 | 说明 |
|---|---|
| 帖子正文 | 截断至 1200 字 |
| 帖子作者的公开显示名 | 用于辅助判断是否为品牌官方账号 |
| 话题标签 | 公开可见的 hashtag |
| 是否含站外链接 | 布尔值 |
| 平台名 | twitter / xiaohongshu / weibo |
我们的服务器把这些内容转发给 TypeSafe AI 的 Jev 模型做分类，取回结果返回给扩展。
当你点击角标上报一条判断错误时，我们会保存：判定结果、你认为正确的类别、
该次判断的概率值、你使用的灵敏度档位和平台名。
上报面板里有一个默认勾选的选项「同时上报这条帖子的文字」。勾选时会一并保存
该帖的正文（截断至 600 字）和作者的公开显示名，用于复核这次判断为何出错；
取消勾选则只保存上述判定数据，不保存任何帖子内容。
上报是完全自愿的行为 —— 不点击就不会发送任何东西。上报数据保存一年。
缓存中不含帖子原文，也不含任何用户标识 —— 同一条热门帖被很多人刷到时，
只需判定一次。
该标识符由扩展随机生成，与你的身份、账号、硬件无任何关联。
帖子原文在返回判定结果后即从内存中丢弃，不写入任何持久化存储、不写入日志。
判定由 TypeSafe AI 的 Jev 模型完成。根据其公开文档，
TypeSafe 不使用客户请求或响应训练模型。详见其
隐私政策。
支付由 Creem 处理，它作为 merchant of record 承担收款、开票与退款。
我们不接触、也不保存你的银行卡信息。
| 权限 | 用途 |
|---|---|
storage | 保存你的设置和本地判定缓存 |
| 三个网站的主机权限 | 读取页面上的帖子文本以进行判定 |
| 服务器域名的主机权限 | 把待判定文本发送到我们的服务器 |
本扩展不申请 tabs、history、cookies、webRequest 等权限，
也不使用 <all_urls>。
邮箱：jevadguide@gmail.com