Scritto 言录 — Privacy Policy / 隐私政策
Effective date / 生效日期: 2026-10-05
Scritto 言录(曾用名 BetterTranscribe;bundle com.jason.BetterTranscribe)
English
Overview
Scritto (言录) is a local-first recording, transcription, and translation app for Mac, iPhone and iPad. It is designed so that everything works entirely on your device by default. By default we do not collect, store, or sell any of your data. The app contains no analytics, no advertising, and no trackers. The one exception is the optional Developer-Sponsored Cloud in TestFlight builds (see below): once you allow it, its gateway keeps device identifiers, usage counters and, if you choose Apple sign-in, a hashed account identifier and login sessions as described below.
What stays on your device
By default, all of the following happens on-device using Apple frameworks, and never leaves your device: audio recording (microphone and system audio), transcription (Apple SpeechAnalyzer), translation (Apple Translation), speaker identification (on-device models), and text polishing (Apple on-device intelligence models). Transcripts and metadata are stored on your device and synced through your own iCloud. Original recordings stay on the device that recorded them; a compressed playback copy of each recording is kept in your own iCloud Drive so your other devices can play it (see below). If you are not signed in to iCloud, or iCloud Drive is off, everything stays local.
iCloud sync
If you are signed in to iCloud, your session library (text and metadata) syncs through your personal, private iCloud database (CloudKit). This data is protected by your Apple Account and is not accessible to the developer or anyone else.
Audio is handled differently: a compressed playback copy of each recording is kept in this app's folder in your iCloud Drive, so every device can play it back. Original recording files stay on the device that recorded them, unless you choose in Settings → General → Recording to move originals to iCloud Drive after 30 days or after 1 year (voice samples recorded to enrol a person are also kept in iCloud Drive). That folder is visible to you in the Files app and in Finder, you can delete files there yourself, and the audio counts against your iCloud storage. This is still your own iCloud — the developer has no access to it.
Small items sync through iCloud key-value storage, which is also part of your own iCloud: your cloud-service configurations (never your API keys), term lists, per-session display preferences, markers that let your other devices remove files you deleted, and your cloud-data permissions (see below).
Optional cloud engines (off by default)
You may optionally configure third-party cloud engines using your own API keys (for example: Deepgram for transcription, pyannoteAI for speaker identification, or LLM providers for translation). These are off by default and only used when you explicitly enable them.
- When you enable a cloud engine, the relevant audio or text is sent directly from your device to that provider to perform the task you requested, under that provider's own privacy policy. Depending on which engines you configure this can include: whole recordings (cloud transcription; speaker identification via pyannoteAI, including when a voiceprint is created or rebuilt; multimodal transcript polishing), and transcript text (cloud translation, text polishing, name inference — which also sends the names already in your people library as a candidate list). Every such action states what leaves the device at the point where you trigger it.
- Cloud-data permissions follow your Apple Account. Before a cloud service first receives your data, the app asks you to allow that service and those types of data. Your permission — and withdrawing it — syncs through your iCloud key-value storage to every device signed in to the same Apple Account, so you are asked once rather than on each device. You are asked again when you add a service address or a new type of data, or when this policy changes in a way that needs your consent again. If iCloud is unavailable, a permission applies only to that device until iCloud is available again. Choosing “Local only” or “Not Now” affects only the device where you chose it. You can review and withdraw permissions in Settings → Cloud Services → Cloud Data Permissions.
- Your API keys are stored in the system Keychain. Keychain items sync through iCloud Keychain, so a key you enter on one device becomes available on your other devices.
- With your own keys, the developer never receives, proxies, or has access to this data.
Developer-Sponsored Cloud (TestFlight only, off until you allow it)
TestFlight builds of Scritto can install a set of ready-to-use cloud configurations (“Sponsored Cloud transcription / document / voiceprint”). You do not enter any API key: requests go from your device to a gateway operated by the developer (a Cloudflare Worker), which forwards them with the developer’s own credentials. Nothing is sent until you have read the recipients listed below and tapped “allow” in the app. That permission follows your Apple Account like any other cloud-data permission (see above), and you can withdraw it at any time in Settings → Cloud Services → Cloud Data Permissions. Switching back to a local engine simply stops using these configurations and does not by itself withdraw the permission. App Store builds do not install these configurations.
- Default routing (“Enhanced Cloud Service”). In the ordinary engine pickers each task offers only Apple (on-device) or “Enhanced Cloud Service”. The Enhanced Cloud Service uses one fixed provider per task: ElevenLabs for transcription and live captions, Alibaba Cloud Model Studio (Bailian) for translation (Qwen-MT), summaries and titles, and for transcript cleanup that includes the matching recording excerpts, and pyannoteAI for voiceprints. The other providers below are used only if you turn on “Show advanced configurations” and pick one of them yourself.
- Recipients. Depending on the configuration you pick, your data reaches one of these providers: Deepgram, AssemblyAI, ElevenLabs, Groq and three SiliconFlow configurations (recording audio, for transcription); one further SiliconFlow configuration (document text, for translation, cleanup, summary and titles); pyannoteAI (recording audio, for speaker identification and voiceprints); Alibaba Cloud Model Studio (Bailian), Google Gemini, Volcengine Ark (Doubao) (document text, for translation, cleanup, summary and titles; two of the Bailian configurations and the Volcengine one also receive the matching recording excerpts whenever you choose them as the transcript-cleanup engine — there is no separate switch; the app says so beneath that setting). The app shows the provider of each configuration in its details, and the permission screen lists all of them.
- What the gateway keeps. The gateway stores no audio and no text. It keeps a device identifier (a random app-generated ID stored in your device keychain, hashed on the server), daily usage counters, the first and most recent day each device was seen, asynchronous job identifiers for up to 48 hours to authorize access to your results, and request status codes. If you choose Sign in with Apple for enhanced cloud services (introduced in beta build 198), the gateway verifies an Apple-signed assertion and stores a hash of its app-specific account identifier, hashes of device-bound session tokens, and account/device quota links. We do not request your name or email. Apple assertions and raw session tokens are not retained or logged; login tokens stay in the local device-only Keychain, not iCloud Keychain, and are not sent to model providers. Daily account/device counters and quota links are kept for 7 days; login challenges expire after 5 minutes and sessions after 30 days unless revoked earlier. Monthly aggregate totals are kept for 3 full months, a device identifier is deleted after 90 days without a metered request, and daily aggregates without account/device identifiers are kept for 400 days. Cleanup runs automatically every day. Your IP address is used only for short-term rate limiting and is not forwarded to providers. Signed-in usage is associated with that cloud account, not anonymous. Local features do not require sign-in; legacy beta clients remain supported during rollout. Settings → Cloud services lets you sign out or delete cloud login data across devices. Deletion does not delete your Apple Account, local documents or recordings, or reset today's allowance; anti-abuse usage records remain for up to 7 days.
- What the providers keep. Each provider retains what its own policy allows; Scritto does not use enterprise zero-retention plans. Providers never receive your Apple Account or any account information. Names in your people library (including the “Me” entry if you renamed it) are sent in two cases: when you re-transcribe with a term list attached (as term hints, to the transcription provider behind that configuration), and when you use smart speaker naming (as a candidate list, to the document provider behind that configuration); the app says so in the configuration details and in the naming confirmation.
- Quota. Each device has a daily quota, and there is a total daily/monthly cap for all testers. When it is used up, or when the service is paused, the app tells you and your local engines keep working. Live translation may switch to Apple’s on-device translation for the rest of that recording, and the app says so on screen; nothing is ever switched to another cloud service.
Calendar access (optional)
If you grant calendar access, the app reads event names for the time of your recording, only to suggest a session name. This information stays on your device.
Data retention and deletion
All data is under your control. Deleting a session inside the app removes its database record, its audio (local and the copy in iCloud Drive), its derived audio, its recovery snapshots and its system-search entry. Deleting a person removes their voiceprints and the enrolment audio. Derived caches on your other devices are cleaned up the next time the app starts there.
Deleting the app removes local data; audio already in iCloud Drive can also be deleted directly in the Files app or in your iCloud settings. What we cannot delete for you: anything a third-party cloud engine (including the providers behind Developer-Sponsored Cloud) kept on its own servers — that is governed by that provider's policy (pyannoteAI, for example, states it deletes uploaded audio after 48 hours).
Children
Apart from the optional cloud identifiers, login sessions and usage counters described above, the app does not collect data from anyone. It is not directed at children.
Changes
If this policy changes, the updated version will be posted at this page with a new effective date. If a change requires your consent again — for example, a new recipient or a new type of data — the app asks you again before sending anything under the new terms.
Contact
Questions: elricedward639@gmail.com
中文
概述
Scritto 言录是一款本地优先的录音、转录与翻译 App(Mac / iPhone / iPad)。默认状态下一切功能完全在你的设备上完成。默认情况下开发者不收集、不存储、不出售你的任何数据。App 不含统计分析、广告和任何追踪器。唯一的例外是 TestFlight 内测版里可选的开发者赞助云(见下):在你允许之后,其网关会保留设备标识与用量计数;若你选择通过 Apple 登录,还会保留账号标识的哈希与登录会话,具体见下。
留在设备上的内容
默认情况下,以下处理全部通过 Apple 框架在设备端完成,数据不离开设备:录音(麦克风与系统音频)、转录(Apple SpeechAnalyzer)、翻译(Apple Translation)、说话人识别(端上模型)、文本整理(Apple 端上智能模型)。文稿与元数据存在设备本地并经你自己的 iCloud 同步;录音原件留在录制它的那台设备上;每场录音另有一份压缩的回放副本放在你自己的 iCloud Drive,供你的其他设备播放(见下)。未登录 iCloud 或关闭 iCloud Drive 时,全部留在本地。
iCloud 同步
若你登录了 iCloud,会话库(文字与元数据)通过你个人的私有 iCloud 数据库(CloudKit)同步。这些数据受你的 Apple 账户保护,开发者与任何第三方均无法访问。
录音音频另有安排:每场录音的压缩回放副本放在你的 iCloud Drive 里本 App 的文件夹,让每台设备都能回放。录音原件留在录制它的那台设备上,除非你在 设置 → 通用 → 录音 中选择 30 天后或 1 年后把原件存到 iCloud Drive(为认人注册而录的声音样本也放在 iCloud Drive)。这个文件夹你在「文件」App 和 Finder 里看得见、可以自己删,音频也会占用你的 iCloud 存储空间。它仍然是你自己的 iCloud,开发者无法访问。
少量小数据经 iCloud 键值存储同步,它同样属于你自己的 iCloud:云服务配置(不含 API key)、术语表、按会话的显示偏好、让其他设备删掉你已删除文件的标记,以及云服务数据许可(见下)。
可选云端引擎(默认关闭)
你可以选择用自己的 API key 配置第三方云端引擎(如 Deepgram 转录、pyannoteAI 声纹识别、LLM 服务商翻译)。它们默认关闭,仅在你显式启用后使用。
- 启用后,相关音频或文本会从你的设备直接发送给该服务商以完成你请求的任务,适用其自身隐私政策。按你配置的引擎不同,可能包括:整段录音音频(云端转录;经 pyannoteAI 做认人,以及每一次生成或重建声纹时;多模态文稿整理),以及文稿文本(云端翻译、文本整理、智能命名——后者还会把你人员库里已有的姓名作为候选名单一并发送)。每一处这样的操作都会在你触发它的地方写明发出去的是什么。
- 云服务数据许可跟着你的 Apple 账户走。某个云服务第一次收到你的数据之前,App 会请你允许该服务处理这些类型的数据。你的许可——以及撤销——经 iCloud 键值存储同步到登录同一 Apple 账户的每台设备,所以只问一次,不必每台设备各问一遍。以下情况会再问:你添加新的服务地址或新的数据类型,或本政策发生需要重新征得同意的变更。iCloud 不可用时,许可暂时只在那台设备生效,iCloud 恢复后再同步。选「仅本地」或「暂不」只影响你做选择的那台设备。可在 设置 → 云服务 → 云服务数据许可 中查看和撤销。
- API key 保存在系统钥匙串中。钥匙串条目会经 iCloud 钥匙串同步,因此在一台设备填过的 key 在你的其他设备上也能用。
- 使用你自己的 key 时,开发者不经手、不代理、也无法访问这些数据。
开发者赞助云(仅 TestFlight 内测版;你允许之前不发送)
Scritto 的 TestFlight 内测版会装入一组现成的云端配置(「赞助云转录 / 文稿 / 声纹」)。你不需要填任何 API key:请求从你的设备发到由开发者运营的网关(一个 Cloudflare Worker),由网关用开发者自己的凭据转发。在你读过下面列出的接收方并在 App 内点「允许」之前,不会发送任何内容。这项许可和其他云服务数据许可一样跟着你的 Apple 账户走(见上),你可随时在 设置 → 云服务 → 云服务数据许可 中撤销对 Scritto 赞助云的许可。改选本地引擎只是不再使用这些配置,本身不撤销许可。App Store 正式版不会装入这些配置。
- 默认路由(「增强云服务」)。普通的引擎选择器里,每项任务只有 Apple(本地)与「增强云服务」两个选项。增强云服务按任务固定使用一家:转录与实时字幕用 ElevenLabs;翻译(Qwen-MT)、总结与标题,以及需要一并发送对应录音片段的文稿整理,用阿里云百炼;声纹用 pyannoteAI。下面其余服务商只在你打开「显示进阶配置」并亲自选中时才会用到。
- 接收方。按你选择的配置不同,数据会发到以下服务商之一:Deepgram、AssemblyAI、ElevenLabs、Groq 及硅基流动的三条配置(录音音频,用于转录);硅基流动另有一条配置(文稿文本,用于翻译、整理、总结与标题);pyannoteAI(录音音频,用于认人与声纹);阿里云百炼、Google Gemini、火山引擎(豆包)(文稿文本,用于翻译、整理、总结与标题;百炼的两条配置和火山那条一旦被选为「文稿整理」引擎,整理时会把对应录音片段一并发送——没有单独的开关,App 在该设置下方有提示)。每条配置的详情里写明其服务商,许可页会把它们全部列出。
- 网关保留什么。网关不存储音频与文本。它保留设备标识(App 随机生成、存在本机钥匙串,服务端存哈希)、按日用量、设备首次与最近使用日期、最多 48 小时的异步任务编号(用于授权读取自己的结果)及请求状态码。若你选择增强云服务的“通过 Apple 登录”(从内测 build 198 提供),网关验证 Apple 签名证明,并保存仅限本 App 的账号标识哈希、本机绑定的登录令牌哈希与账号/设备用量关联。不请求姓名或邮箱;Apple 签名证明与原始令牌不落盘、不写日志。登录令牌仅存本机 deviceOnly 钥匙串,不经 iCloud 钥匙串同步,也不传给模型服务商。账号/设备的按日计数与额度关联保留 7 天;登录挑战五分钟失效,会话至撤销或 30 天届满。按月全站合计保留 3 个整月;设备标识在 90 天无计量请求后删除;不含账号/设备标识的每日全站汇总保留 400 天。每天自动清理。IP 地址仅用于短时限速,不转发给服务商。登录后的用量关联云服务账号,不再称为匿名。本地功能无需登录;迁移期间旧内测版仍可使用。设置 → 云服务可退出,或删除此账号全部设备上的云登录资料;不会删除 Apple 账号、本地文稿或录音,不会重置当日额度,防重复计量记录最多保留七天。
- 服务商保留什么。各家按其自身政策保留;Scritto 未使用企业版零留存方案。服务商不会收到你的 Apple 账号或任何账号信息。人员库中的姓名(若你把「我」改成了真名,也包括它)会在两种情况下随请求发送:挂载术语表后重新转录(作为术语提示,发给该配置背后的转录服务商),以及智能认人(作为候选名单,发给该配置背后的文稿服务商);App 在配置详情与智能认人确认框中说明这一点。
- 额度。每台设备每天有额度,所有内测者合计也有日/月上限。额度用完或服务暂停时,App 会明确告知,你的本地引擎照常可用。实时翻译可能在这场录音剩下的时间里改用 Apple 本地翻译,屏幕上会写明;绝不会改发到别的云端服务。
日历权限(可选)
若你授权日历,App 仅读取录音时段对应的事件名,用于推荐会话名称。该信息不离开设备。
数据保留与删除
所有数据由你掌控。在 App 内删除一条会话,会一并删除它的数据库记录、音频(本地以及 iCloud Drive 里那份)、派生音频、重处理前的恢复副本,以及系统搜索里的条目。删除一个人,会一并删除 TA 的声纹与注册录音原件。你其他设备上的派生缓存会在那台设备下次启动 App 时清理。
删除 App 会清除设备本地数据;已经在 iCloud Drive 里的录音也可以直接在「文件」App 或系统 iCloud 设置里删除。我们替你删不掉的:第三方云端引擎(含开发者赞助云背后的服务商)留在它自己服务器上的内容——那由该服务商的政策决定(例如 pyannoteAI 声明上传音频 48 小时后删除)。
儿童
除上文所述可选云服务的标识、登录会话与用量计数外,本 App 不向任何人收集数据;本 App 不面向儿童。
变更
政策如有变更,将在本页更新并标注新的生效日期。若变更需要重新征得你的同意(例如新增接收方或新的数据类型),App 会在按新条款发送任何数据之前再次询问。
联系方式
问题反馈:elricedward639@gmail.com