数据与隐私
更新于 2026-10-03 · 适用于文脉在线试用版
阅读资料默认存在当前浏览器。你可以选择把文章和问答放到账户云端档案,换设备继续。使用 AI 时,相关文字会发送给模型服务,无论是否开启云端存储。
保存在浏览器的内容
文章正文、来源网址、问答及对应段落、笔记、阅读位置、概念记录和界面设置保存在当前浏览器。仅登录不会自动上传旧资料。原来的本机档案与账户档案分别保存,切换不会删除原记录。清除站点数据会删除本机内容,包括尚未同步的改动。
可选的云端档案
在设置中选择文章并同意上传后,文章全文、来源、阅读进度、问题来路、答案、笔记和未发送草稿会保存到文脉使用的 Cloudflare D1 数据库。进入云端档案后,新文章和后续改动也会自动上传。你可以另行勾选同步伴读方法,其中包含反馈原话和回答摘录,即使未选择对应文章;新设备不会自动开启伴读学习。
跨篇统计、显示偏好、密码和模型密钥不在档案同步范围内。云端通过登录账户隔离访问,但不是端到端加密。本机缓存也未单独加密,退出登录不会擦除磁盘上的副本,请勿在公用设备存放私密材料。
两端新增的问题可以合并;冲突的草稿或回答会保留两边,等你处理。阅读和输入过程中不会自动替换当前内容,其他设备的更新需在设置中选择“同步并接续”。当前容量上限为账户合计 10 MB、200 条记录(含删除标记与伴读方法),每篇文章及问答 1.2 MB;超出时保留本机并提示,不会悄悄丢弃。
使用文脉额度无需提供密钥。也可在“账户与设置 → 模型服务”主动接入自己的智谱 API Key。密钥以 AES-GCM 加密后存入服务器数据库,解密密钥由服务器单独保存;这不是端到端加密,服务器调用模型时会解密使用。界面只返回末四位,密钥不进入阅读档案、备份或普通调用日志。自带密钥由服务商核算用量,失败时不会自动改扣文脉额度。
可在模型服务页移除服务器上的个人密钥,后续调用将改用文脉额度;已发出的请求不会因此撤回。清除浏览器数据、退出登录或删除云端阅读档案不会删除该密钥,需在模型服务页另行移除。如需立即撤销密钥本身,请同时在智谱平台撤销。旧版曾保存的本地密钥不会自动上传,可能仍在本地设置里。
保存在服务器的内容
除你选择同步的档案外,以下记录存于文脉使用的 Cloudflare 数据库,用于登录、计算试用次数和排查故障。
- 账户:账户名、状态、创建时间、试用次数和到期时间。
- 密码:注册、登录时会发送给文脉服务器。数据库保存用于验证密码的加盐哈希和盐值,不存明文。
- 登录记录:凭证哈希、创建时间、到期时间和最近使用时间。浏览器用 Cookie 保持登录,页面脚本不能读取该 Cookie。登录最长有效 30 天,与 AI 试用期限不同。
- 注册限制:将 IP 地址和服务器密钥一起做哈希,用处理后的标识限制同一网络重复注册。这不等于完全匿名,Cloudflare 仍会接触原始 IP。
- AI 用量:所属账户、功能、模型、服务来源、文章标识、请求时间、结果状态、模型返回的用量、错误代码和是否退还平台额度。用量页展示最近 30 天;这不是后台记录的自动删除期限。
注册不要求邮箱、手机号或付款信息。官网没有广告或行为分析脚本。
哪些操作会发送内容
- 网址导入:网址发给文脉服务器,由服务器提取正文。请勿提交带有私密访问凭证的网址。
- 网页捕手:点击“提取并送到文脉”后,扩展会尝试展开、滚动当前页面,再读取标题、正文和来源网址。提取结果暂存在扩展中,5 分钟后失效;传给文脉后,扩展中的副本会删除。扩展不读取或导出网站 Cookie。
- AI 功能:预读、提问、调整回答或检查理解时,相关原文、问题和必要的对话会经文脉服务器发送给智谱模型服务。模型密钥由服务器保存,浏览器不直接调用模型服务。
- 事实核验:待核验内容可能发送给模型或检索服务。
账户和用量记录不用于保存文章正文或完整问答。Cloudflare 和模型服务商可能保留请求记录。请勿导入机密或敏感材料。
备份和删除
设置中的“个人档案”可选择文章,备份原文、问答来路、草稿、阅读进度和伴读方法;选择全部文章时可包含跨篇记录,也可包含说话方式。JSON 文件可重新导入且未加密,不含模型密钥或登录凭证,请妥善保管。
云端档案中删除文章,会同步删除云端正文和阅读记录,保留用于防止旧设备复活资料的删除标记。暂停同步只影响此设备,既不删除本机副本,也不删除云端。设置中的“删除云端档案”会移除当前账户的在线档案并使旧同步版本失效;所有设备的本机副本仍在,不会重新自动上传。
Cloudflare 的灾备副本可能在其恢复保留期内继续存在,最长 30 天,参见 D1 备份说明。云端档案不设自动到期删除;试用 AI 到期不锁住已有档案和导出。云端不是备份的替代品,重要内容请另存文件。
本机档案的清空功能会删除自行导入的文章及问答,并重置样例和阅读进度;跨篇记录和显示设置会保留。要删除此浏览器中的全部档案缓存和旧副本,请在浏览器中清除站点数据。此操作不会替你删除云端档案。
退出登录会使当前登录失效。清空浏览器数据会删除本地资料和登录 Cookie,但这两种操作都不会删除服务器上的账户或用量记录。
目前没有自助注销或删除服务器账户和用量记录的入口,也未设置这些记录的自动删除期限。删除云端档案不等于注销账户,试用到期也不代表记录已删除。
功能和数据处理方式可能调整。目前没有正式客服或密码找回功能。
Data & privacy
Updated October 3, 2026 · Applies to the online trial
Reading data stays in this browser by default. You can choose an account cloud archive to continue on another device. AI features send relevant text to the model service whether or not cloud storage is enabled.
What stays in your browser
Article text, source URLs, anchored Q&A, notes, reading positions, concept records and interface settings are saved in this browser. Signing in alone does not upload existing data. Your original device archive and account archives are stored separately; switching does not delete them. Clearing site data removes device content, including unsynced changes.
Optional cloud archive
After you select articles and consent in Settings, their full text, sources, progress, question paths, answers, notes and unsent drafts are stored in Wenmai's Cloudflare D1 database. New articles and later changes within the cloud archive upload automatically. You may separately opt in to companion-method sync, including feedback and answer excerpts even from unselected articles. Learning does not turn on automatically on new devices.
Cross-article statistics, display preferences, passwords and model keys are not synced as archive content. Cloud access is account-isolated but not end-to-end encrypted. Browser caches are not separately encrypted; signing out does not erase disk copies. Avoid private material on shared devices.
Independent questions can merge. Conflicting drafts or answers retain both versions for review. Reading and typing are not replaced in the background; choose “Sync and continue” in Settings to load updates. Current limits are 10 MB and 200 records per account, including deletion markers and methods, and 1.2 MB per article with Q&A. Overflow remains on-device with a notice.
No key is needed to use Wenmai’s allowance. You may connect your own Zhipu API key in Account & settings → Model service. It is stored on the server using AES-GCM encryption with a separate server-held encryption key. This is not end-to-end encryption: the server decrypts the key when calling the model. Only its last four characters are returned to the interface. Keys are excluded from reading archives, backups and ordinary request logs. Your provider accounts for BYOK usage; failed calls never silently switch to Wenmai’s allowance.
Remove a saved key on the model service page to switch future requests to Wenmai’s allowance. Requests already sent are not recalled. Clearing browser data, signing out or deleting a cloud reading archive does not remove this key. To revoke the key itself immediately, also revoke it at Zhipu. Legacy browser-stored keys are not uploaded automatically and may still exist in local settings.
What is stored on the server
In addition to opted-in archives, these records are stored in Wenmai's Cloudflare database for sign-in, trial limits and troubleshooting.
- Account: username, status, creation time, trial allowance, and expiry.
- Password: sent to Wenmai's server when you register or sign in. The database stores a salted password hash and salt, not the plain-text password.
- Sessions: credential hash, creation time, expiry, and last use. A cookie keeps you signed in and cannot be read by page scripts. Sessions last up to 30 days, separately from the AI trial period.
- Registration limits: your IP address is hashed with a server secret to limit repeat sign-ups from the same network. This is not full anonymity; Cloudflare still receives the original IP address.
- AI usage: account, feature, model, service source, article identifier, request time, result status, reported tokens, error codes and platform refunds. The usage page shows the last 30 days; this is not an automatic deletion deadline for server records.
Registration does not require an email address, phone number, or payment details. The website has no advertising or behavioral analytics scripts.
When content is sent
- Importing a URL: the URL goes to Wenmai's server to extract the article. Do not submit URLs containing private access credentials.
- Web Catcher: when you choose “Extract and send to Wenmai”, the extension tries to expand and scroll the current page, then reads its title, article text, and source URL. Captures are stored temporarily in the extension, expire after five minutes, and are removed from the extension after transfer to Wenmai. The extension does not read or export website cookies.
- AI features: previewing, asking questions, adjusting answers, or checking understanding sends relevant source text, questions, and necessary conversation through Wenmai's server to Zhipu's model service. Model keys stay on the server; the browser does not call the model service directly.
- Fact-checking: claims may be sent to a model or search service.
Account and usage records are not used to store article bodies or full conversations. Cloudflare and model providers may retain request records. Do not import confidential or sensitive material.
Backups and deletion
“Personal archive” in Settings exports selected articles, question paths, drafts, progress and companion methods as restorable JSON. Selecting all articles also allows cross-article records; speaking preferences are optional. Files are unencrypted and exclude model keys and login credentials. Keep them safe.
Deleting an article in a cloud archive removes its cloud text and reading records, leaving a deletion marker to prevent old devices from restoring it. Pausing sync affects this device only and erases neither copy. “Erase cloud archive” removes the account's live archive and invalidates old sync versions. Device copies remain and do not automatically upload again.
Cloudflare recovery copies may remain for up to 30 days; see D1 backups. Cloud archives have no automatic expiry. AI trial expiry does not lock existing archives or exports. Sync is not a substitute for backups; export important content.
The device archive's clear-data action removes imported articles and their Q&A and resets samples and progress; cross-article records and display settings remain. Clearing this site's data in your browser removes all device archive caches and previous copies, not the cloud archive.
Signing out invalidates the current session. Clearing browser data removes local records and the sign-in cookie. Neither action deletes server-side account or usage records.
There is currently no self-service deletion of server-side account or usage records, and no automatic retention deadline for them. Erasing a cloud archive does not delete your account; trial expiry does not erase records.
Features and data handling may change. There is currently no formal customer support or password recovery.