Pro Note Taker / 速记达人

Privacy Policy

Last updated: November 1, 2026 · Applies to Pro Note Taker 1.0.4 for macOS, iOS and iPadOS

Pro Note Taker is a Mac app for live transcription, system-audio capture, media import, translation, subtitles, and AI summaries. This policy explains what Pro Note Taker processes, what stays on your Mac, exactly when content leaves your device, and the choices you have.

iPhone and iPad

The mobile app records microphone audio and imports files you select. Its large captions appear only inside the app. System-audio capture, cross-app floating subtitles and the optional AI Search/MCP server described below are Mac-only features.

On iPhone and iPad, records, transcripts, translations, notes, import/recovery files and preferences stay in this app’s local container. API keys stay in the device Keychain. Selected cloud services receive the same audio or text and authentication data described below, after service-specific consent. Apple Speech may use Apple servers. Apple Translation uses downloaded language packs; Apple Intelligence notes run on-device when available, otherwise the app clearly labels its extractive key-point preview. We operate no processing server, app account system, advertising or analytics service.

Ollama on mobile connects over HTTPS to a computer or server you operate. Processing takes place on that server, not on the iPhone or iPad. Local-network access is used only if you choose a local-network server address. No Mac recordings or API keys are automatically copied to the mobile app.

Imported media is copied into app-owned storage so interrupted jobs can be retried. Local import/recovery data has no automatic expiry; deleting the app removes its local container. You can delete saved records in Library, remove keys and revoke cloud permissions in Settings. Your original files outside the app are not deleted. Exports are shared only through the destination you choose. Cloud providers’ retention terms apply to data sent to them.

The detailed provider disclosures, cloud processing information and contact details below also apply to mobile; references to “your Mac” mean your selected device except for the explicitly Mac-only features above. Mac workflow details are retained for Mac users.

1. The short version

2. What Pro Note Taker can access

Depending on the features you use, Pro Note Taker may access:

3. What stays on your Mac

Apple Speech is not listed as local-only because the Speech framework may send audio to Apple servers. Pro Note Taker asks for Apple Speech cloud-processing consent before use.

4. Cloud providers and exact data flows

Pro Note Taker sends content off your Mac only to the service selected for that feature and only after showing consent for that specific service. BYOK providers use your own API key.

Apple Speech — optional transcription

If you select Apple Speech, microphone, system-audio, or imported audio is processed through Apple's Speech framework. Recognition may use Apple servers, so Pro Note Taker requires explicit Apple Speech cloud-processing consent before starting it.

Alibaba Cloud Bailian / DashScope — optional transcription (ASR)

When Bailian is selected, live microphone or system audio is sent as a continuous PCM stream over an encrypted WebSocket (wss) to the selected DashScope endpoint, using fun-asr-realtime by default. Imported files are converted locally into 60-second AAC chunks (WAV if compression fails) and sent as base64 data URIs in HTTPS requests, using fun-asr-flash-2026-06-15 by default.

Alibaba Cloud Bailian / Qwen — optional summaries

When you select Bailian as the summary provider, transcript text is sent to the DashScope chat API and processed by Qwen models (qwen3.8-flash by default; qwen3.8-max for the "Exhaustive Notes" style).

Deepgram — optional transcription

If you select Deepgram, live audio is sent as a continuous PCM stream over an encrypted WebSocket to api.deepgram.com, using Nova-3. Imported audio and recovery requests use HTTPS file requests to the same provider. After a live connection failure, retained audio may be sent to Deepgram within the current session to recover missing results, subject to your Deepgram consent. Deepgram is currently the engine in Pro Note Taker that provides speaker diarization (speaker labels).

DeepL — optional translation

If you select DeepL, transcript text is sent to DeepL's API for translation.

OpenAI and Anthropic / Claude — optional AI notes

When selected, transcript text and summarization instructions are sent directly to api.openai.com or api.anthropic.com, using your API key and chosen model. Audio files are not sent through these summary connections. OpenAI requests set store: false; provider security retention and account policies still apply. We do not control a provider's retention or training settings.

Ollama and compatible AI servers

Ollama sends transcript text to your configured Ollama endpoint. With a local model running on this Mac, inference stays on your Mac. Ollama must be running and the model must already be installed; the app does not download models. Cloud models and models reported by Ollama as hosted remotely are not supported through the local Ollama connection.

A compatible server receives transcript text, instructions, the model ID, and your API key if configured. We cannot determine how a custom server processes or retains data. A remote Ollama or compatible connection requires consent that displays the destination. Changing the endpoint or summary provider resets custom-server consent. HTTPS is required except for HTTP connections to loopback addresses on this Mac.

Load Models requests only a model list and sends no recording or transcript. It may send your API key to the selected service to authenticate.

Each provider's own privacy policy governs what they do with the requests you send them:

5. Bring Your Own Key (BYOK)

Bailian, Deepgram, DeepL, OpenAI, Anthropic, and compatible AI services use API keys that you create yourself with each provider. Keys are stored exclusively in the macOS Keychain on your device. Apple Speech does not use an API key. Pro Note Taker never proxies cloud-processing requests through any server of ours — requests go directly from your Mac to the selected provider.

6. Cloud-processing consent

Pro Note Taker records consent separately for Apple Speech, Bailian transcription, Deepgram transcription, DeepL translation, Bailian/Qwen summaries, OpenAI, Anthropic, and configured AI servers. The consent sheet identifies the service and the type of content that will leave your Mac. You can revoke all cloud-processing consent at any time in Settings; Pro Note Taker will ask again before the next cloud request. The privacy policy is also available from Settings → General.

7. AI search for other apps

Settings → AI Search can let another app on this Mac read saved meetings through a local MCP port. It is off until you turn it on. The other app must present a token that stays in your Mac’s Keychain. By default the port accepts connections only from this Mac.

If you also turn on access for other computers on the same network, a program on another computer can read those meetings when it has the token. Pro Note Taker does not upload meetings to its own server. You can turn the feature off or rotate the token at any time.

8. No account, no tracking

Pro Note Taker does not include a login system, advertising SDKs, in-app behavioral tracking, third-party analytics SDKs, or third-party crash-reporting SDKs. We do not sell personal information.

9. Diagnostic logging

App Store Release builds do not create debug.log, do not emit Pro Note Taker diagnostic logs, and do not retain transcript fragments for debugging.

Development Debug builds may keep local diagnostic logs, including short transcript or translation fragments. They are reset on launch and are not automatically uploaded. Logs leave your Mac only if you choose to share them. Recovery audio and transcript drafts are separate local data used to recover your work; they are not diagnostic uploads.

10. Permissions

Each permission is used only for the feature you run.

11. Data retention and deletion

Transcripts, recording metadata, and settings stay on your Mac until you delete them. To recover interrupted work, supported live workflows also keep a local PCM audio journal and periodic transcript drafts in the app's Application Support directory. Live drafts are saved about every two seconds and again when recording stops; the newest words may therefore not yet be in the last draft after a crash.

After a complete transcript is successfully saved to the local database, its temporary recovery audio and draft are removed. If recognition, translation, or saving is incomplete, recovery data is kept. Partial or recovered records retain their associated audio for review; deleting the record also attempts to remove that app-owned recovery audio, and reports a cleanup failure if it cannot. The app does not delete your original imported files.

The import queue stores the chosen file's location, a read-only security-scoped bookmark, provider and language choices, status, partial text, and completed audio-chunk progress. Reopening the app restores this queue without automatically uploading or restarting paid work; you must choose Retry. A completed job clears its saved text and bookmark after the record is saved; Clear Finished removes its remaining queue entry. Failed jobs retain recovery information. Recovery data has no automatic expiry.

You can delete records inside the app, remove saved API keys in Settings, or remove the app's container and local support files when uninstalling. For content sent to cloud providers, their retention policies apply.

12. Cloud processing

Cloud features send content only to the provider you select, and only after you give consent.

13. Children

Pro Note Taker is not directed at children under 13.

14. Changes

We may update this policy; material changes will update the "Last updated" date above.

15. Contact

Privacy and support questions: ftchftch@gmail.com

隐私政策

最后更新:2026-11-01 · 适用于速记达人 1.0.4(macOS、iOS、iPadOS)

速记达人是一款 Mac 应用,提供实时转写、系统音频录制、媒体导入、翻译、悬浮字幕和 AI 摘要。本政策说明速记达人处理哪些数据、哪些数据留在本机、内容何时离开你的设备,以及你有哪些选择。

iPhone 与 iPad 版

手机版录制麦克风音频,并导入你主动选择的文件。大字字幕只在应用内显示。下文的系统音频捕获、跨 App 悬浮字幕与可选 AI 检索/MCP 服务器仅属于 Mac 版。

iPhone 和 iPad 上的记录、转写、译文、笔记、导入与恢复文件及设置保存在应用本机容器;API Key 保存在设备钥匙串。云端处理经所选服务的专门同意后,发送下文列明的音频或文字与身份验证数据。Apple Speech 可能使用 Apple 服务器;Apple 翻译使用已下载语言包;可用时 Apple 智能笔记在设备端生成,否则明确标注本地要点提取预览。我们没有处理服务器、应用账户系统、广告或分析服务。

手机版 Ollama 通过 HTTPS 连接你运行的电脑或服务器,推理在该服务器完成,而非手机端。仅在你填写局域网服务器地址时使用本地网络权限。Mac 上的录音和 API Key 不会自动复制到手机。

导入媒体会复制到应用自有目录,方便中断后重试;本机导入与恢复数据没有自动到期时间。删除应用可移除其本机容器。你可在资料库删除记录,在设置移除密钥或撤销云端权限;应用不会删除原始外部文件。导出文件仅经你选择的目标分享。已经发送给云服务商的数据遵循其保留政策。

下方服务商披露、云端处理与联系方式同样适用手机版;除明确标注的 Mac 专用功能外,“Mac”指你正在使用的设备。为 Mac 用户保留下方 Mac 工作流程细节。

1. 要点速览

2. 速记达人可以访问什么

取决于你使用的功能,速记达人可能访问:

3. 留在本机的数据

Apple Speech 未列为纯本地服务,因为 Speech 框架可能将音频发送至 Apple 服务器。速记达人会在使用前请求 Apple Speech 云端处理同意。

4. 云服务与确切的数据流向

速记达人只会把内容发送给你为该功能选择的服务,并且只在针对该具体服务展示同意提示后进行。BYOK 服务使用你自己的 API Key。

Apple Speech —— 可选转写

选择 Apple Speech 后,麦克风、系统音频或导入音频会通过 Apple Speech 框架处理。识别可能使用 Apple 服务器,因此速记达人会在启动前请求明确的 Apple Speech 云端处理同意。

阿里云百炼 / DashScope —— 可选转写(ASR)

选择百炼后,实时麦克风或系统音频以连续 PCM 流通过加密 WebSocket(wss)发送至所选 DashScope 端点,默认使用 fun-asr-realtime。导入文件在本机转换为 60 秒的 AAC 分段(压缩失败时使用 WAV),再以 base64 数据 URI 通过 HTTPS 请求发送,默认使用 fun-asr-flash-2026-06-15。

阿里云百炼 / 通义千问 Qwen —— 可选摘要

当摘要服务商为百炼时(默认为百炼),转写文本会发送至 DashScope 对话 API,由 Qwen 模型处理(默认 qwen3.8-flash;"详尽笔记"样式使用 qwen3.8-max)。

Deepgram —— 可选转写

选择 Deepgram 后,实时音频以连续 PCM 流通过加密 WebSocket 发送至 api.deepgram.com,使用 Nova-3。导入音频及恢复请求通过 HTTPS 文件请求发送至同一服务商。实时连接失败后,在你的 Deepgram 同意有效的前提下,当前会话可能将保留音频发送给 Deepgram,以恢复缺失结果。Deepgram 是目前速记达人中提供说话人区分(说话人标签)的引擎。

DeepL —— 可选翻译

选择 DeepL 后,转写文本会发送至 DeepL API 进行翻译。

各服务商的隐私政策约束其对你所发送请求的处理方式:

OpenAI 与 Anthropic / Claude —— 可选 AI 笔记

选择后,转录文字和总结指令会直接发送至 api.openai.com 或 api.anthropic.com,使用你的 API Key 和所选模型。这些总结连接不会发送音频文件。OpenAI 请求设置 store: false,但服务商的安全留存及账户政策仍适用。我们不控制服务商的留存和训练设置。

Ollama 与兼容 AI 服务

Ollama 将转录文字发送至你配置的 Ollama 地址。使用运行于本机的本地模型时,推理在 Mac 上完成。需要自行安装模型并保持 Ollama 运行;本应用不下载模型。本地 Ollama 连接不支持云端模型,也不支持被 Ollama 标记为远程托管的模型。

兼容服务会收到转录文字、指令、模型 ID 以及你配置的 API Key。自定义服务如何处理或保留数据取决于其运营方。远程 Ollama 或兼容服务连接会显示目标地址并征求同意;修改地址或切换总结服务会重置自定义服务的同意记录。除本机回环地址可用 HTTP 外,其他地址必须使用 HTTPS。

「加载模型」只请求模型列表,不发送录音或转录文字;可能向所选服务发送你的 API Key 以验证身份。

5. 自带密钥(BYOK)

百炼、Deepgram、DeepL、OpenAI、Anthropic 和兼容 AI 服务 使用你自行向服务商申请的 API Key。密钥仅保存在你设备的 macOS 钥匙串中。Apple Speech 不使用 API Key。速记达人不会经由我们的任何服务器中转云端处理请求——请求直接从你的 Mac 发往所选服务商。

6. 云端处理同意

速记达人分别记录 Apple Speech、百炼转写、Deepgram 转写、DeepL 翻译、百炼/Qwen 摘要、OpenAI、Anthropic 及配置的 AI 服务的同意状态。同意页会说明服务名称及将离开 Mac 的内容类型。你可随时在设置中撤销所有云端处理同意;下次发起云端请求前,速记达人会再次询问。隐私政策也可以从「设置 → 通用」打开。

7. 给其他 App 的 AI 检索

「设置 → AI 检索」可以让这台 Mac 上的其他 App 通过本机 MCP 端口读取已保存的会议。这项功能默认关闭,打开后对方必须提供保存在钥匙串中的令牌。默认只接受来自这台 Mac 的连接。

如果你再打开「允许同一网络上的其他电脑连接」,同一网络里拿到令牌的程序也可以读取这些会议。速记达人不会把会议上传到自己的服务器。你可以随时关闭这项功能,或更换令牌。

8. 无账号、无追踪

速记达人不包含登录系统、广告 SDK、行为追踪、第三方分析 SDK 或第三方崩溃上报 SDK。我们不出售个人信息。

9. 诊断日志

App Store Release 构建不会创建 debug.log,不会输出速记达人诊断日志,也不会为调试保留转写片段。

开发用 Debug 构建可能保留本机诊断日志,其中可能包含简短的转写或翻译片段。日志在启动时重置,不会自动上传;只有你主动分享时才会离开 Mac。恢复音频和转写草稿是用于找回工作的独立本机数据,不属于诊断上传。

10. 权限

每项权限仅用于你主动运行的功能。

11. 数据保留与删除

转写、录音元数据和设置在你删除之前一直保留在本机。为恢复被中断的工作,支持的实时流程还会在应用的 Application Support 目录保留本机 PCM 音频恢复文件和定期转写草稿。实时草稿约每两秒保存一次,停止录音时再次保存;因此崩溃前最新的文字可能尚未进入最后一份草稿。

完整转写成功保存到本机数据库后,对应的临时恢复音频和草稿会被清理。识别、翻译或保存未完成时,恢复数据会保留。部分或恢复记录会保留关联音频供检查;删除记录也会尝试删除应用持有的该份恢复音频,清理失败会显示错误。应用不会删除你原始的导入文件。

导入队列保存所选文件位置、只读安全作用域书签、服务商和语言选择、状态、部分文本及已完成音频块进度。重新打开应用只恢复队列,不会自动上传或重启付费任务;你需要点击「重试」。完成的任务在记录保存后清空队列中的文本和书签,「清除已完成」会移除其余队列条目。失败任务保留恢复信息。恢复数据没有自动过期时间。

你可以在 App 内删除记录、在设置中移除已保存的 API Key,或在卸载时移除应用容器及本机支持文件。发送至云端服务商的内容适用其各自的保留政策。

12. 云端处理

云端功能只会在你同意后,把内容发送给你选择的服务商处理。

13. 儿童

速记达人不面向 13 岁以下儿童。

14. 政策变更

我们可能不时更新本政策;重大变更会更新上方的"最后更新"日期。

15. 联系我们

隐私与支持问题:ftchftch@gmail.com