# Privacy Notice / 隐私声明

> Source: https://listen.365ui.com/privacy
>
> Effective date: 2026-11-22

## Privacy Notice

Effective date: 2026-11-22

### 1. Who is responsible and what this covers

This notice is published by 365UI (operated by Yu Chen as a sole proprietorship; "we", "us") and covers the preparation, pricing (including the enterprise inquiry form), problem report, sign-in, account (including monthly captions), order, client, AV, audio-feed, attendee and on-device live captions pages at listen.365ui.com, the iOS app “365UI Interpretation Audio” (in Chinese, 365UI 同传接声), and the interpretation service on the event day. Privacy contact: Yu Chen, support@365ui.com.

### 2. Information we collect

- Accounts: the email address you sign up or sign in with; when you use Google, the account identifier, name and Google-verified email address Google provides; the name and organization you enter on your account page; whether you want product news and offers; when you signed up, when you last signed in and how; and your answer to the “How did you hear about us?” question asked once after you sign in, with the line you add (you can skip it; a skip is recorded too, and you are not asked again). For a reseller account we open, also the reseller’s name, phone number, sign-in email addresses (there can be several), GST/HST registration number (if any), and which version of the reseller terms it accepted and when. We do not record which website you came from. Signing in stores a session cookie in your browser that only keeps you signed in (30 days); signing in with Google also uses a 10-minute cookie that guards against forged requests.
- Ordering: the event name, times and rooms; where the event takes place (state, province or country, to work out sales tax); the on-site contact’s name, email address and mobile number; a referral code, if entered; and, for an order placed while signed in, the account it belongs to. For an order placed through a reseller, the reseller gives us the client’s on-site contact details.
- Enterprise inquiries: the name, work email, company or organization, optional event dates and size, and message you enter on the pricing page. They reach us by email and are not stored in the order system; we send no automatic confirmation to the address you enter.
- Problem reports: on the “Report a problem” page, the kind of problem you choose, what you write, an optional email and up to three screenshots, with the browser and system version your browser reports and the page language at the time; when you come from one of our apps’ menus, the app’s version as well. Screenshots are made smaller in your browser before they are sent. The report and screenshots reach us by email and are not stored in the order system; we send no automatic confirmation to the address you enter. Please cover other people’s information in a screenshot first.
- Payment: handled by Stripe; we never see full card numbers, and when you pay for prepaid credit from a US bank account (ACH), Stripe handles the bank verification and we never see full bank account numbers either. If you choose Alipay, the payment is completed on Alipay’s page or in its app, and Alipay handles it under its own privacy policy. Stripe receives the contact email and the billing details you enter on its payment page (billing name, company name, billing address and tax IDs), and passes us the payment status, transaction identifiers and those billing details. We keep them, with any purchase order number you enter when ordering or buying prepaid credit, to issue your invoice. You can correct an event’s billing details on its client page after paying.
- Client and AV pages: names and mobile numbers of on-site helpers and venue audio staff; speakers’ names, titles and organizations; questionnaire answers; the name of the person filling in the form, if given; the organizer and sponsor logos you upload (PNG or JPEG images); requests to cancel or reschedule, and remote help bought, on the client page, with any note; when each room’s audio-feed link was first opened successfully (no IP address or device); questions typed under “Ask a question” on the client page (used only to write that answer; we keep neither the question nor the answer); and questions to the AI assistant on the audio-feed page, which go to the model that answers them with the page’s and the room’s status at that moment (which step the page is at, which device the sound comes from, how loud we hear it, how good the laptop’s connection is, whether it is plugged in and its battery level, whether we receive it, whether the interpretation engines work, how many listen), and which we keep, with that status and the event and room, for 90 days to improve the answers and to look into problems on the day.
- Event materials: the agenda, program book, speaker bios and slides or scripts you send us (PDF, PPTX, DOCX or TXT files), used to prepare the event’s glossary.
- On the event day: the room audio (every microphone feeding the mixer, audience-question microphones included; or what the mics worn by speakers pick up), transcripts of the original speech and the interpretation, and captions.
- The on-device live captions page: the sound from the microphone or audio interface is recognized as text in this computer’s browser, then translated into the other language by the browser’s own on-device translation (Chrome, Edge). The sound, the original text and the translation stay on this computer; they are sent neither to us nor to any recognition or translation service. You sign in to an account before you start (see “Accounts” above); we only read whether the account has bought the paid version, and record neither how long you use the page nor what it recognizes. When you buy the paid version, we keep how many months it was, when it was paid and when it ends. The first time you start, the browser downloads the recognition model file from Oracle Cloud’s public storage; the language packs for translation are downloaded by the browser itself, handled under its maker’s terms and shared by all websites.
- Monthly captions: when you buy monthly captions on your account page, the two caption languages and time zone you choose, any purchase order number, when the plan was paid and when it ends, the standing attendee code and standing audio-feed link we issue for it (the link stored only as a hash), and each session you book under it (kept as a captions-only online event, with its times, any precheck minutes used and any minutes returned to that period when a session was unavailable on our side).
- The iOS app “365UI Interpretation Audio”: when a helper uses it in place of the audio-feed page, the app sends the sound of the chosen microphone into that room’s interpretation, where it is handled as sound from the audio-feed page is. The app also sends the code in the audio-feed link (exchanged for the credential that lets it into that room) and the app’s version; the chosen input’s name, channel count and sample rate; a status once a second (checking, sending, input lost, on a break); an identifier generated at random when the app is installed, which tells phones apart in our logs and is not tied to a person; and the phone’s model and system version (sent to LiveKit by the connection component). The app needs no account and reads no contacts, location or photos; the camera is used only to scan the audio-feed QR code, and its picture is neither kept nor uploaded; its notifications are raised on the phone itself (to tell the helper that a phone conversation or another app has taken the microphone, or that the mic’s receiver was unplugged) and do not pass through our servers. On the phone the app keeps the language choice, that identifier and a diagnostic record: times, states, level figures, the names of the chosen input and the room, and, when the connection has trouble, the connection component’s own log (which holds the phone’s network addresses and the room’s identifier); the record holds no sound, no audio-feed code and no room credential. The app never uploads the record; it leaves the phone only if the helper shares it or copies it off, and it is deleted with the app. The app also keeps the audio-feed code and the name of the one room it last got into in the phone’s system keychain (on this phone only, in no backup and synced to no other device), so that after the app is closed one press brings it back to that room; the app deletes it once the link is replaced or the event has ended, and on its first launch after being deleted and installed again. “Try a microphone first” in the app only measures the level on the phone: it uses no network and sends no sound.
- Technical information: IP addresses, browser and connection logs produced when someone visits the pages or joins a room; at ordering we use the IP address (for IPv6, its prefix) to limit how often orders can be placed. When a page fails, your browser sends us the error message, the page address with its private parts removed, the site version and a browser type (such as “WeChat on iOS”); the report we log leaves out your IP address. We count page views with Vercel Web Analytics, and how often three things happen: a tap on “See 365UI” on a listener page, a free trial starting and a payment going through (only the event’s name, whether the visit came from a listener page, and whether the payment was by card, prepaid credit or a reseller); once a tab, it also records the browser’s first language (only the language code, such as es or ko, Chinese split into Traditional and Simplified, no region) and whether the tab opened on an event page or the rest of the site, to see which languages the site should add; it uses no cookies; page addresses are reported with event names, access codes and link keys removed. Conversations in the “AI assistant” at the bottom right of the site’s pages: each question goes to the model that answers it with up to ten earlier questions and answers of the same conversation, so a follow-up is understood; we keep each question, its answer and up to 3 questions before it for 90 days, to see which questions are answered badly and to improve the answers and the site, without the IP address and unlinked to any account or order (the assistant says so where the conversation starts); we limit how often one IP address can ask, and keep nothing past that limit.
- Browser storage: the language choice, an unsent order draft, a referral code a partner’s link brought in (kept for 60 days), whether this tab came from a listener page, whether this tab’s language has been counted, and the conversation in the “AI assistant” (all three gone when the tab closes); the audio-feed page also keeps a randomly generated device id, the audio interface or mic and input chosen for each room, and that room’s audio-feed link (which stops working when it expires); the on-device live captions page keeps the recognition model file in this browser’s cache, so the next start needs no download, and clearing this site’s data removes it. We do not use advertising cookies or track people across other websites.

### 3. How we use it

To price and manage orders, take payment and issue invoices; to run your account, so your account page can list your orders; to learn how new customers heard of us (only from the answer you choose), so we can improve how we describe the service; to prepare and deliver interpretation (the speaker list, and the glossary prepared from any event materials you provide, are sent to the interpretation engine), the highlights and refined translations after the event, and, where the client turns it on, the “what was said so far” recap; to communicate about the event (confirmation, sign-in and notice emails, preparation reminders 7 days and 1 day before, a reminder 7 days before monthly captions end, notices when transcripts and refined translations are ready, and one request for feedback after the event; we quote your feedback only if you clearly agree in your reply); to reply to your inquiry; to read the conversations with the “AI assistant” and the audio-feed page’s assistant, to improve the answers, the site and the guides and to look into problems on the day; to send product news and offers if you ticked the box for them, which you can turn off on your account page at any time; to secure the service, fix faults and review quality; and to comply with the law and keep necessary business records. We do not sell personal information or event content, and we do not use them to train our own models.

### 4. Where it goes

- Oracle Cloud (United States; the primary server is in San Jose, California, and the backup server may be in another US region): the servers that process the live audio; for rooms with voice and a listener limit of 900 or more, the servers that deliver interpreted audio and captions to listeners (San Jose, with a backup in Phoenix, Arizona; not recorded, not kept after delivery); the archive of audio and transcripts, and the nightly encrypted backups of our order database; the on-device live captions page’s recognition model file is in its public storage too, and it receives the browser’s IP address when the browser downloads it.
- Our own model server (United States): runs Kimi K3 to answer the questions typed under “Ask a question” on the client page, to the audio-feed page’s assistant, or into the “AI assistant” at the bottom right of the site’s pages, with up to ten earlier questions and answers of an “AI assistant” conversation, the audio-feed page’s status when it asks, and our guides and site pages. The server keeps no copy of the questions or answers (we keep the “AI assistant” and audio-feed page conversations as section 2 says).
- Alibaba Cloud Model Studio (Singapore): primary interpretation; it also receives the speaker list and, when you provide event materials, the text we extract from them, to find the names, organizations and terms for the event’s glossary; for an event with listening languages other than Chinese and English, or whose client tells us a speaker speaks a language that is not a listening language, the entries of the event’s glossary, to translate them into those languages for the interpretation; before the event, the terms of the event’s glossary (not names or organizations), to write test sentences, synthesize them and run them through the interpretation, to see which need rewording, with the results shown only on your event page; when you send terms on the free trial page to test them before ordering, those terms (not names or organizations) in the same way, with the results shown only on your trial page; for a room with the speaker-labeled transcript ordered, that room’s recording after the event for transcription (held in its temporary storage and deleted after 48 hours); for events ordered online that keep transcripts, each room’s captions in every language after the event, to write the highlights; for a room with the speaker-labeled transcript ordered, the transcript’s text (without the speaker labels), to translate it again with care into each listening language (and into any other language the client asks for afterwards); for a room where the client turned on “what was said so far”, the captions so far, about every ten minutes during the event, to write the points; for a room where the client chose to imitate each speaker’s own voice, it learns each speaker’s voice from the live audio to speak the translation in it (we create no voice profile for any speaker; the OpenAI backup does not do this); and, when our own model server (the item above) is unavailable or we switch to this one, the questions listed there, to write the answers. Under its terms, API inputs and outputs are kept for up to 30 days and are not used to train models without separate consent.
- OpenAI (United States): backup interpretation, used when the primary has problems. We have turned off sharing data with OpenAI; under its API terms the data is not used to train models, and abuse-monitoring logs are kept for up to 30 days.
- Zoom, Microsoft (Teams) or Cisco (Webex): when the helper on the audio-feed page or you on the client page choose to send captions into the event’s Zoom, Teams or Webex meeting, that room’s captions in one language go to that meeting, where that provider handles them under the terms and settings of the meeting host’s (or organizer’s) account. The caption address the meeting gives is held only in the browser page and in our servers’ memory, never logged or written to disk, and is dropped when the room closes. When the meeting host chooses to authorize with their own Zoom account on the audio-feed page or the client page, we read only that meeting’s caption address from Zoom and revoke the authorization right after; we keep neither the token Zoom issues nor any account details.
- Recall.ai (United States): for a room set to an online meeting, it sends our bot “365UI 同传 · Live interpretation” into that Zoom, Google Meet or Microsoft Teams meeting; the bot’s video is a page of ours, which sends the meeting’s sound to interpretation. As a participant the bot sees and hears the meeting, its participant list and its chat, and it receives the room’s meeting link (with its passcode), which it clears by itself a few days later. We have turned its video and recording off, and anything recorded is deleted within an hour. When it joins, the bot posts a notice and the listening link in the meeting chat.
- LiveKit Cloud: transport and delivery of live audio and captions (for rooms with voice and a listener limit of 900 or more, our servers on Oracle Cloud deliver to listeners instead, while the venue sound still goes through LiveKit Cloud). We have not turned on its recording, so audio and captions are not kept after delivery; operational data such as connection records is kept for up to 60 days.
- Cloudflare (United States): for listeners who only read the captions, its Workers relay the caption text to their phones as it comes. It holds the room’s last 30 lines in memory only, clears them once the room has had no captions for a while and writes nothing to storage; operational data such as connection records is kept under its terms.
- Vercel (United States): website hosting, which keeps access logs (IP addresses and the addresses visited, including client event links) for up to 30 days. Error reports from pages (without IP addresses) are kept in its function logs, also for up to 30 days. The site uses its Web Analytics to count page views, the three things above and browsers’ first languages: no cookies, and addresses are reported with their private parts removed.
- Upstash (eastern United States): accounts, sign-in sessions, orders, questionnaire answers, event records, the conversations with the site’s “AI assistant” and the audio-feed page’s assistant, and the billing details and sales records behind invoices; for events ordered online, the transcript copies the client downloads (speaker-labeled and refined ones included), the highlights and their publishing settings (the organizer’s link included), the “what was said so far” points, and each room’s listener counts (numbers only, no listener identities); the logo images you upload; the paid version of on-device live captions, how long each purchase was and when it ends; monthly captions languages, time zones, purchases and session ledgers; the terms you send for a test on the free trial page and the results (deleted 30 days after you send them); and order, trial, sign-in and inquiry rate counters by IP address (or its network), account and email (they expire within an hour to a day).
- Resend (United States): confirmation, sign-in and notice emails, and forwarding enterprise inquiries and problem reports (with their screenshots) to us.
- Google (United States): if you choose to sign in with Google, Google confirms who you are and gives us your email address, name and account identifier; Google handles that sign-in under its own privacy policy.
- Stripe (United States): payments.
- Apple (United States): the iOS app is installed from the App Store or TestFlight. Apple handles the download records, the crash reports and usage figures you agree to share with developers, and any feedback you send in TestFlight under its own privacy policy, and passes them to us; none of it holds venue audio.
- Referrers: when an order uses a referrer’s code (including a customer’s own code from their account page), we tell that referrer only the order date, the amount and the commission or reward credit it earns, to settle it; we do not share the event name or contact details.
- Resellers: for an order placed through a reseller, the reseller receives copies of the event’s operational emails (reminders, system alerts on the day, transcript and highlights notices, the follow-up after the event) and, on its partner page, sees the client’s on-site contact details and preparation progress.
- Cloudflare, Google and Resend (United States): email sent to support@365ui.com is forwarded by Cloudflare to both our Gmail mailbox and our inbox at Resend, which we reply from.
- Cloudflare (United States): the human check before a free trial starts, when sending terms for a test on the trial page, when requesting a sign-in link, and when sending an enterprise inquiry or a problem report (Turnstile). It reads signals from your browser and device and your IP address to tell a person from a script; we receive only whether it passed.

We use each provider under its business or API terms; the retention periods above are taken from the terms they published in October 2026, and their current terms prevail. Under those terms a provider may keep logs or copies to run its service, for security or to monitor abuse, and such copies are not covered by our own deletion periods below. Processing takes place in the United States, Singapore and other countries the providers use, where courts, law enforcement or national-security authorities may lawfully obtain information. We may also disclose information when the law requires it, or to protect our rights and resolve disputes.

### 5. How long we keep it

- Accounts: kept until you ask us to delete them, with the account’s purchases of the paid version of on-device live captions and its monthly captions ledger (each session’s event record is still deleted 30 days after the session ends, under event records below). Sign-in links stop working after 20 minutes, and sign-in sessions after 30 days or as soon as you sign out.
- Unpaid orders: deleted within 4 days.
- Event records, the answers on the client and AV pages, uploaded logo images, the downloadable transcript copies, the highlights and listener counts: deleted 30 days after the event ends, when a published highlights page goes offline too. The “what was said so far” points: kept for at most 3 days, or deleted with the event record if that expires sooner.
- Events whose organizer chose to keep the transcripts (in the questionnaire, or on the event page after the event): the event record (with the answers on the client and AV pages), the downloadable transcript copies (each language’s transcript, the speaker-labeled transcript, the refined translations), the highlights and listener counts are kept until the organizer changes it back on the event page or asks us to delete them, and a published highlights page stays online; changed back, they are deleted on the original date, or a day later if that date has passed. This is the organizer’s choice and instruction, and telling speakers and attendees is the organizer’s responsibility. Uploaded logo images, the venue recordings and the archive of audio and transcripts below are still deleted on schedule.
- The archive of audio and transcripts: kept 30 days by default, then deleted; if you choose not to keep it, we make no archive, though the live processing still takes place. For an event that keeps recordings, after the event we copy each room’s venue recordings from its slots to Oracle Cloud object storage in San Jose for you to download from your event page (each download link stops working an hour after you click it); the copies are deleted 30 days after they are made.
- Backups of our order database (everything listed under Upstash above): one each night, encrypted, readable only with a key that is not kept on our servers; each backup, including any copy pulled to our own computer, is deleted after 14 days. So anything that expires, or that we delete at your request, under the items above normally disappears from the backups within a further 14 days (if backups keep failing, the last good one is kept until the next succeeds); if we restore from a backup, each record still expires on its original schedule.
- Conversations with the site’s “AI assistant” and the audio-feed page’s assistant: deleted after 90 days (an audio-feed page’s names its event and room, and is kept that long even when the event record goes sooner).
- Local copies of answers we export to prepare an event: deleted 30 days after the event ends.
- Event materials you send us: the text we extract from them is used only while preparing the glossary and is not stored separately; the files, and the report from preparing the glossary, are kept on our own computer and deleted 30 days after the event ends; materials sent by email are kept with that email as correspondence, below.
- Names prepared from a speaker list or event materials: used only for that client’s later events; the client may ask us to delete them.
- The record behind a standing attendee sign: only a code, an account identifier computed one-way from the email address, and the room name, with no email address or name; renewed each time the code is scanned or the sign is printed, kept 400 days after the last use, and deleted at once when the code is replaced.
- Payment and accounting records (including the billing details on invoices), and correspondence (including enterprise inquiries and problem reports): kept for the periods tax and accounting rules require, or as long as needed to resolve a dispute.

### 6. Your choices and rights

- When signed in, you can change your name, organization and whether you get product news and offers on your account page.
- Until the event ends, you can see and change answers on the client and AV pages, and choose not to keep the archive; after the event, the organizer can also choose on the event page to keep the transcripts, or to have them deleted on schedule again.
- Highlights stay private until the organizer reviews them on the event page and publishes them; the organizer can take out any item or unpublish at any time. Once published, anyone who opens the room’s address can read them, and the page asks search engines not to index it. The summary and key points leave out the names on the speaker list; a quote may carry a speaker’s name, from the speaker list and only where the speaker-labeled transcript recognizes them. A speaker who wants their words taken out can ask the organizer, or write to us. “What was said so far” exists only if the client turns it on and is not published: during the event only someone who opens the listener page with the room’s access code can read it; its points likewise leave out the names on the speaker list and carry no quotes, and the page says it is machine-written and may be wrong.
- Write to support@365ui.com to ask to see, correct or delete your personal information early, or to challenge our practices. We may need to verify your identity; records the law requires us to keep are excepted; separating one person’s voice from a multi-speaker recording may need the organizer’s help.

### 7. Security

We take reasonable measures suited to the sensitivity of the information (for example, client, AV and sign-in links and sign-in sessions are stored only as hashes, and logs leave out transcript text), but no Internet service can be completely secure. If we confirm a security incident involving personal information, we will notify the client, affected individuals and regulators as and when the law requires. Client and AV links work as credentials; share them only with those who need them.

### 8. Sensitive information and children

The service is not for health information that identifies a patient, identity-document or financial account numbers, or other sensitive information, and is not directed at children under 13; see section 9 of the terms of service.

### 9. Do Not Track signals

We do not track people across other websites, so we do not treat browser Do Not Track signals differently.

### 10. Changes

When this notice changes, we update its version date on this page; material changes apply only to orders placed after the change. This notice is written in Chinese and English; if the two differ, the English version prevails.

### 11. Contact

support@365ui.com

## 隐私声明

生效日期：2026-11-22

### 1. 谁负责、适用于什么

本声明由 365UI（经营者：Yu Chen，个人独资经营；以下称「我们」）发布，适用于 listen.365ui.com 上的准备页、价格页（含企业咨询表单）、「报告问题」页、登录页、账户页（含包月字幕）、下单页、客户专属页、AV 页、接声页、听众页、本机同传字幕页，iOS App「365UI 同传接声」（英文名 365UI Interpretation Audio），以及活动当天的同传服务。隐私负责人：Yu Chen，联系邮箱 support@365ui.com。

### 2. 我们收集的信息

- 账户：注册或登录用的邮箱；用 Google 登录时，Google 提供的账户编号、姓名和经 Google 验证的邮箱；您在账户页填写的姓名和机构；是否愿意接收产品更新和优惠；注册时间、最近一次登录的时间和登录方式；登录后页面上问一次的「您是怎么知道我们的」，您选的那一项和补充的一句话（可以跳过，跳过也会记下，之后不再问）。我们开通的代理账户，另有代理的名称、电话、登录邮箱（可以有几个）、GST/HST 登记号（如果有），以及同意代理条款的版本和时间。我们不记录您是从哪个网站点过来的。登录后浏览器会保存一个只用于保持登录的会话 Cookie（30 天）；用 Google 登录时还会用一个 10 分钟的临时 Cookie 防止伪造请求。
- 下单：活动名称、时间和房间；活动所在地（州、省或国家，用于确定销售税）；现场联系人的姓名、邮箱和手机号；推荐码（如果填了）；登录状态下下单时，订单所属的账户。通过代理下的订单，客户现场联系人的信息由代理交给我们。
- 企业咨询：您在价格页留言时填写的姓名、工作邮箱、公司或机构、（选填的）活动日期和规模，以及留言内容。内容以邮件发给我们，不存入订单系统；我们不会给您填写的邮箱自动发确认信。
- 问题报告：您在「报告问题」页选的问题类别、写的经过、（选填的）邮箱和最多三张截图，以及发送时您的浏览器自己报告的浏览器和系统版本、页面语言；从我们的 App 菜单点过来时，还有 App 的版本号。截图在您的浏览器里缩小后发出。内容连同截图以邮件发给我们，不存入订单系统；我们不会给您填写的邮箱自动发确认信。截图里如果有别人的信息，请先遮掉。
- 付款：由 Stripe 处理，我们看不到完整卡号；选用美国银行账户（ACH）购买预付额度时，银行账户验证由 Stripe 处理，我们同样看不到完整银行账号。选用支付宝付款时，付款在支付宝的页面或 App 里完成，由支付宝按它自己的隐私政策处理。Stripe 会收到联系人邮箱和您在付款页填写的开票信息（账单姓名、公司名称、账单地址、税号），并把付款状态、交易编号和这些开票信息转给我们；我们保存它们，连同您下单或购买预付额度时填写的采购单号，用于开具发票。活动的开票信息付款后可以在专属页面修改。
- 专属页面和 AV 页：现场帮手和场地音响人员的姓名和手机号；讲者的姓名、职衔和机构；问卷答案；填写人主动留下的姓名；您上传的主办方标识和赞助商标识（PNG 或 JPEG 图片）；在专属页面提交的取消或改期申请、远程协助的购买和留言；每个房间的接声链接第一次成功打开的时间（不记 IP 地址和设备）；在专属页面「有问题直接问」里输入的问题（只用来生成这一次的回答，我们不保存问题和回答）；在接声页问 AI 助手的问题，会连同提问那一刻接声页和这个房间的状态（接声页在哪一步、声音从哪个设备来、远程听到的电平、这台电脑的网络质量、是否插着电源和电量、系统收没收到声音、翻译引擎是否正常、有几位听众）一起发给回答的模型，我们把问题、回答和这些状态连同活动和房间保存 90 天，用来改进回答和排查现场问题。
- 活动资料：您发给我们的议程、程序册、讲者介绍和讲稿（PDF、PPTX、DOCX 或 TXT 文件），用来整理本场的专有名词表。
- 活动当天：会场声音（接入调音台的所有话筒，包括观众提问话筒；或讲者戴的话筒收到的声音）、原文和译文的转写、字幕。
- 本机同传字幕页：话筒或声卡的声音在这台电脑的浏览器里识别成文字，再由浏览器自带的本机翻译（Chrome、Edge）译成另一种语言。声音、原文和译文都留在这台电脑上，不发给我们，也不发给任何识别或翻译服务。开始之前要登录账户（见上面「账户」一项）；我们只读这个账户有没有买付费版，不记录您在页面上用了多久、识别了什么。买付费版时，我们保存购买的月数、付款时间和到期时间。第一次开始时，浏览器从 Oracle Cloud 的公开存储下载识别模型文件；翻译用的语言包由浏览器自己下载，按浏览器厂商的条款处理，所有网站共用。
- 包月字幕：您在账户页购买包月字幕时选定的两种字幕语言、所在时区、（选填的）采购单号、付款和到期时间，我们为包月生成的长期观众码和长期接声链接（链接只存哈希值），以及您在包月内预订的每一场场次（作为一场只出字幕的在线下单活动保存，含时段、预检用掉的分钟数和因我们一方原因无法收听而退回当期的分钟数）。
- iOS App「365UI 同传接声」：帮手用它代替接声页时，App 把所选话筒的声音送进这间房的同传，之后的处理和接声页送来的声音相同。App 另外发送：接声链接里的接声码（用来换取进入这间房的凭证）和 App 的版本号；所选输入的名字、声道数和采样率；每秒一次的状态（声检中、推送中、话筒不在、茶歇）；一个安装时随机生成的编号，用来在我们的日志里分辨是哪一台手机，不和人关联；手机的型号和系统版本（由连接组件发给 LiveKit）。App 不要账号，不读通讯录、位置和照片；相机只用来扫接声二维码，画面不保存也不上传；通知只在手机本机发出（话筒被电话或别的 App 占用、或话筒接收器被拔掉时提醒帮手），不经过我们的服务器。App 在手机上保存语言选择、那个编号和诊断记录：时间、状态、电平数字、所选输入和房间的名字，以及连接出问题时连接组件自己的日志（里面有手机的网络地址和房间编号）；记录里没有声音、接声码和进入房间的凭证。诊断记录不会自己上传，只有帮手自己把它分享出去或从手机上拷走才会离开手机，删除 App 时一并删除。App 还把最近进过的那一个房间的接声码和房间名存在这台手机的系统钥匙串里（只留在这台手机上，不进备份，也不同步到别的设备），这样 App 被关掉后点一下就能回到那间房；链接换过或活动结束后 App 会删掉它，删除 App 后重装，第一次打开时也会删掉。App 里的「先试一下话筒」只在手机上量电平，不联网，不送出任何声音。
- 技术信息：访问页面或接入房间时产生的 IP 地址、浏览器和连接日志；下单时我们用 IP 地址（IPv6 取前缀）限制下单频率。页面出错时，浏览器会把错误信息、去掉私密部分的页面地址、网站版本和浏览器类型（如「微信 iOS」）发给我们，我们记下的报告不含 IP 地址。我们用 Vercel Web Analytics 统计页面浏览量，以及三件事各发生了几次：在收听页点「看看 365UI」、开始免费试用、付款完成（只记事件名、是不是从收听页点进来的、付款用的是银行卡、预付额度还是代理）；每个标签页还记一次浏览器的首选语言（只记语言代码，比如 es、ko，中文分繁简，不记地区）和打开的是活动页面还是网站其他页面，用来看网站该加哪些语言；它不使用 Cookie；上报的页面地址先去掉活动编号、访问码和链接里的密钥。网站页面右下角「AI 客服」里的对话：每问一句，这段对话里前面最多十问十答会和它一起发给回答的模型，好让它听懂追问；每一问的问题、回答和它前面最多 3 问，我们保存 90 天，用来查哪些问题答得不好、改进回答和网站内容，不记 IP 地址，也不和账户或订单关联（对话框开头也这样写着）；我们按 IP 地址限制提问的次数，超过次数的问题不保存。
- 浏览器本地存储：语言选择、未提交的订单草稿、推荐链接带来的推荐码（保存 60 天）、这个标签页是不是从收听页点进来的、这个标签页的语言统计过没有和「AI 客服」里这段对话的问答（这三样关掉标签页就没了）；接声页还会存一个随机生成的设备编号、每个房间选择的声卡或话筒和输入，以及该房间的接声链接（链接到期后失效）；本机同传字幕页把识别模型文件存进这个浏览器的缓存，下次开始不用再下载，清除这个网站的数据就会删掉。我们不使用广告 Cookie，也不做跨网站跟踪。

### 3. 我们怎么用

用于：报价、管理订单、收款和开具发票；管理账户，让您在账户页找到自己的订单；了解新客户是怎么知道我们的（只看您自己选的答案），用来改进介绍文字；准备和提供同传（讲者名单，以及从您提供的活动资料里整理出的专有名词表，会发给翻译引擎）、会后精华、会后精翻稿和（客户开了才有的）「前面讲了什么」；活动沟通（确认信、登录链接、通知邮件、活动前 7 天和 1 天的准备提醒、包月字幕到期前 7 天的提醒、文字稿和精翻稿可以下载的通知，以及活动后一封征求意见的邮件；只有您回信明确同意，我们才会引用您的评价）；回复您的企业咨询；查看「AI 客服」和接声页 AI 助手的问答，改进回答、网站内容和使用说明，排查现场问题；在您勾选同意后发送产品更新和优惠，随时可以在账户页取消；保障安全、排查故障和复核质量；遵守法律和保留必要的经营记录。我们不出售个人信息或活动内容，也不用它们训练我们自己的模型。

### 4. 信息去往哪里

- Oracle Cloud（美国；主用服务器在加利福尼亚州圣何塞，备用服务器可能在美国其他区域）：处理现场音频的服务器；收听上限 900 人及以上的有译音房间，把译音和字幕分发给听众的服务器（圣何塞，备用在亚利桑那州凤凰城，不录制，送达后不保存）；原音和转写的归档，以及订单库每晚的加密备份；本机同传字幕页的识别模型文件也放在它的公开存储里，浏览器下载时它会收到浏览器的 IP 地址。
- 我们自己的模型服务器（美国）：运行 Kimi K3，回答专属页面「有问题直接问」、接声页 AI 助手和网站页面右下角「AI 客服」里的问题，连同「AI 客服」这段对话里前面最多十问十答、接声页提问时的状态，以及我们的使用说明和网站页面的文字。这台服务器不另外保存问题和回答（「AI 客服」和接声页的问答我们按第 2 节保存）。
- 阿里云 Model Studio（新加坡）：主用同传，也会收到讲者名单；您提供活动资料时，我们从中抽出的文字会发给它，用来找出人名、机构和术语，整理成本场的专有名词表；订了中英以外收听语言、或者告诉我们有讲者讲收听语言以外的语言的活动，我们还会把本场专有名词表里的词条发给它，译成这些语言，用在同传里；开场前，我们还会把本场专有名词表里的术语（不含人名和机构）发给它，写成测试句、合成语音，再经同传各测一遍，看哪些要改写，结果只放在您的活动页上；您在免费试用页提交术语做下单前的实测时，同样把这些术语（不含人名和机构）发给它，结果只放在您的试用页上；订了会后分人文字稿的房间，活动后还会把这间房的录音文件传给它转写（放在它的临时存储里，48 小时后删除）；保留文字稿的在线下单活动，活动后还会把每间房各语言的字幕文字发给它整理精华；订了会后分人文字稿的房间，还会把分人文字稿的文字（不含说话人标签）发给它精翻成各收听语言（客户事后要补译其他语言时同样发给它）；客户开了「前面讲了什么」的房间，活动进行中还会每十分钟左右把已有的字幕文字发给它整理要点；客户选了「模仿讲者本人的声音」的房间，它从现场的声音里学每位讲者的声音，用来念译文（我们不为讲者另建声音档案；备用的 OpenAI 不做这件事）；我们自己的模型服务器（见上一项）不可用时、或我们切换到它时，那一项里的问题改发给它生成回答。按其条款，接口调用的输入和输出最多保存 30 天，未经另行同意不用于训练模型。
- OpenAI（美国）：备用同传，主用出问题时使用。我们已关闭向 OpenAI 共享数据的选项；按其接口条款不用于训练模型，滥用监测记录最多保存 30 天。
- Zoom、Microsoft（Teams）或 Cisco（Webex）：帮手在接声页或您在专属页面选择把字幕送进这场活动的 Zoom、Teams 或 Webex 会议时，那间房一种语言的字幕文字会发给这场会议，由这家服务商按会议主持人（或组织者）账户的条款和设置处理。会议给的字幕地址只在浏览器页面和我们服务器的内存里，不写日志、不存盘，房间关闭就丢掉。会议主持人选择在接声页或专属页面用自己的 Zoom 账号授权时，我们只向 Zoom 读取这一场会议的字幕地址，读完立即撤销这次授权，不保存 Zoom 给的令牌和账户信息。
- Recall.ai（美国）：房间选了「线上会议」时，它替我们把机器人「365UI 同传 · Live interpretation」派进那场 Zoom、Google Meet 或 Microsoft Teams 会议，机器人的画面是我们的网页，会议的声音由这个网页送进同传。机器人作为一位参会者，看得到、听得到这场会议，也看得到参会者名单和会议聊天；它会拿到这个房间的会议链接（含密码），这个链接由它在几天后自动清除。我们关闭了它的录像和录制，留下的录制材料一小时内删除。机器人进会时在会议聊天里发一条说明和收听链接。
- LiveKit Cloud：实时音频和字幕的传输和分发（收听上限 900 人及以上的有译音房间，听众这一侧改由我们在 Oracle Cloud 的服务器分发，会场送声仍经 LiveKit Cloud）。我们没有开启它的录制，音频和字幕送达后不保存；连接记录等运行数据最多保存 60 天。
- Cloudflare（美国）：只看字幕的听众，字幕文字经它的 Workers 实时转发到手机。它只在内存里留这间房最近 30 句，房间一段时间没有字幕就清空，不写入存储；连接记录等运行数据按它的条款保存。
- Vercel（美国）：网站托管，会记录访问日志（IP 地址和访问的网址，包括客户专属链接），最多保存 30 天。网页出错时的错误报告（不含 IP 地址）记在它的函数日志里，也最多保存 30 天。网站用它的 Web Analytics 统计页面浏览量、上面那三件事的次数和浏览器的首选语言：不使用 Cookie，上报的地址先去掉私密部分。
- Upstash（美国东部）：账户、登录会话、订单、问卷答案、活动记录、网站「AI 客服」和接声页 AI 助手的问答，发票所用的开票信息和销售记录，在线下单的活动还有供客户下载的文字稿副本（含分人文字稿和精翻稿）、会后精华和它的发布设置（含主办方链接）、「前面讲了什么」的摘要，以及每个房间的听众人数（只有人数，不含听众身份），以及您上传的标识图片，本机同传字幕付费版的购买记录和到期时间，包月字幕的语言、时区、购买和场次账本，在免费试用页提交实测的术语和实测结果（提交后 30 天删除），以及按 IP 地址（或它所在的网段）、账户和邮箱计的下单、试用、登录和企业咨询频率计数（1 小时到 1 天内自动过期）。
- Resend（美国）：发送确认信、登录链接和通知邮件，并把企业咨询和问题报告（含截图）转发给我们。
- Google（美国）：您选择用 Google 登录时，Google 核实您的身份，并把您的邮箱、姓名和账户编号告诉我们；Google 按它自己的隐私政策处理这次登录。
- Stripe（美国）：付款。
- Apple（美国）：iOS App 从 App Store 或 TestFlight 安装。下载记录、您同意与开发者共享时的崩溃报告和使用统计，以及您在 TestFlight 里提交的反馈，由 Apple 按它自己的隐私政策处理，再提供给我们；其中没有会场的声音。
- 推荐人：订单用了某位推荐人的推荐码时（包括客户在账户页拿到的自己的推荐码），我们只告诉对方订单日期、金额和这一单的佣金或奖励额度，用于结算；不提供活动名称和联系人信息。
- 代理：通过代理下的单，代理收到这场活动运行邮件（提醒、活动当天的系统告警、文字稿和精华通知、会后回访）的副本，并在它的代理页上看到客户的现场联系人信息和准备进度。
- Cloudflare、Google 和 Resend（美国）：寄到 support@365ui.com 的邮件由 Cloudflare 转发两份，一份到我们的 Gmail 邮箱，一份到我们在 Resend 的收件箱，我们从这个收件箱回信。
- Cloudflare（美国）：免费试用开始前、在试用页提交术语实测、请求登录链接、提交企业咨询和提交问题报告时的人机验证（Turnstile）。它读取您浏览器和设备的信号以及 IP 地址，判断是不是真人在操作；我们只收到通过与否。

我们按各服务商的企业或接口条款使用它们；上面的保存期限摘自它们 2026 年 10 月公开的条款，以它们的最新条款为准。它们可能按这些条款，为提供服务、安全或滥用监测保存日志或副本，这些副本不受下文我们自己的删除期限约束。处理在美国、新加坡和服务商使用的其他国家进行，当地法院、执法或国家安全机关可能依法调取。在法律要求，或为保护我们的权利、处理纠纷时，我们也可能提供信息。

### 5. 保存多久

- 账户：保存到您要求删除为止，本机同传字幕付费版的记录和包月字幕账本随账户保存（包月订出的每一场活动记录仍按下面活动记录那一项在场次结束后 30 天删除）。登录链接 20 分钟后失效，登录会话 30 天后失效，退出登录时立即失效。
- 未付款的订单：4 天内删除。
- 活动记录、问卷和 AV 页的答案、上传的标识图片、供下载的文字稿副本、会后精华和听众人数：活动结束后 30 天删除，已发布的精华页也随之下线。「前面讲了什么」的摘要：最多保存 3 天，活动记录到期更早时随它一起删除。
- 主办方选择「文字稿一直保留」的活动（在问卷里选，或活动结束后在专属页面选）：活动记录（含问卷和 AV 页的答案）、供下载的文字稿副本（各语言文字稿、分人文字稿、精翻稿）、会后精华和听众人数保存到主办方在专属页面改回、或写信要求删除为止，已发布的精华页也一直在线；改回时按原定日期删除，原定日期已过的，一天后删除。这是主办方的选择和指示，对讲者和听众的告知由主办方负责。上传的标识图片、会场录音和下面的原音转写归档仍按期删除。
- 原音和转写归档：默认保存 30 天后删除；选择「不保存」时我们不留归档，但现场的实时处理照常进行。保存录音的活动，会后我们把每个房间时段内的会场录音另复制一份到 Oracle Cloud 圣何塞的对象存储，供您在专属页面下载（每次点下载生成的链接 1 小时后失效），复制后 30 天自动删除。
- 订单库（上面 Upstash 一项里的全部内容）的备份：每晚一份，加密保存，只有一把不放在服务器上的密钥能打开；每份（包括拉到本地的副本）保存 14 天后删除。因此上面各项到期或应您要求删除后，通常再过 14 天也从备份中消失（备份连续失败时，最近一份成功的会留到下一次成功为止）；如果从备份恢复，每条记录仍按原定时间到期。
- 网站「AI 客服」和接声页 AI 助手的问答：保存 90 天后删除（接声页的问答记着活动和房间，活动记录先删了也照样保存到期）。
- 我们为准备活动导出到本地的答案副本：活动结束后 30 天删除。
- 您发给我们的活动资料：从中抽出的文字只在整理专有名词表时使用，不另外保存；资料文件和整理时的报告存在我们自己的电脑上，活动结束后 30 天删除；资料随邮件寄来的，邮件按下面往来邮件那一项保存。
- 按讲者名单或活动资料制作的人名词表：只用于该客户之后的活动，客户可以要求删除。
- 长期观众牌的对应记录：只有一个编号、一个由邮箱单向换算出的账户标识和房间名，不含邮箱或姓名；每次有人扫码或打印时顺延，最后一次使用后保存 400 天，作废时立即删除。
- 付款和账务记录（含发票上的开票信息）、往来邮件（含企业咨询和问题报告）：按税务和会计规定，或处理纠纷所需的期限保存。

### 6. 您的选择和权利

- 登录后可以在账户页修改姓名、机构，以及是否接收产品更新和优惠。
- 活动结束前，您可以在专属页面或 AV 页面查看和修改答案，也可以选择不保存归档；活动结束后，主办方还可以在专属页面选择文字稿一直保留，或改回到期删除。
- 会后精华默认不公开，主办方在专属页面核对后才发布，可以逐条去掉、随时取消发布。发布后任何打开房间地址的人都能看到，页面不让搜索引擎收录。概要和要点不写讲者名单上的人名；金句旁可能标讲者姓名（来自讲者名单，只在分人文字稿认得出时）。讲者想去掉自己的话，可以找主办方，也可以写信给我们。「前面讲了什么」只有客户开了才有，不公开发布：活动进行中，用房间通行码打开听众页的人才能读到；要点同样不写讲者名单上的人名，也没有金句；页面上写明是机器整理、可能有误。
- 写信到 support@365ui.com，可以要求查阅、更正或提前删除您的个人信息，或对我们的做法提出异议。我们可能需要核实身份；法律要求保留的记录除外；在多人录音中分离某个人的声音，可能需要活动主办方协助。

### 7. 安全

我们采取与信息敏感程度相称的合理措施（例如客户链接、AV 链接、登录链接和登录会话只存哈希，日志不写转写文字），但任何互联网服务都不能保证绝对安全。确认发生涉及个人信息的安全事件时，我们会在法律要求的时间和范围内通知客户、受影响的个人和监管机关。客户链接和 AV 链接相当于凭据，请只发给需要的人。

### 8. 敏感信息和儿童

本服务不用于处理可识别患者的健康信息、证件或金融账号等敏感信息，也不面向 13 岁以下儿童；见服务条款第 9 条。

### 9. 「请勿跟踪」信号

我们不做跨网站跟踪，因此不对浏览器的「请勿跟踪」（Do Not Track）信号作特殊处理。

### 10. 变更

本声明有变更时，我们在本页更新版本日期；重大变更只适用于变更之后的订单。本声明有中文和英文两个版本；两者不一致时，以英文版为准。

### 11. 联系我们

support@365ui.com
