English

Privacy

Last updated 26 September 2026 · applies to UTUVO Type for iPhone (app and keyboard), Android and Mac.

Short version: UTUVO Type has no accounts, ads or tracking. We do not operate an analytics service or a server that receives your voice or text. On Android, Google's ML Kit sends diagnostics and usage metrics to Google when used. Audio or text may go directly to the speech or AI provider you use; the routes and controls are listed below.

Your voice

On iPhone, speech uses Apple's Speech framework. Recognition prefers on-device processing when available. You can allow Apple server recognition in Settings → Recognition & Privacy; setting Only use on-device recognition prevents network recognition and refuses unsupported languages. Apple controls the actual route when server recognition is allowed, under Apple's privacy policy. The dictation screen labels whether the request is on-device-only or allows server recognition.
Optional cloud recognition on iPhone (off by default): if you turn on Cloud recognition in Settings → Smart Cleanup, the recording of each dictation is sent with your own API key to the provider you chose there (Google Gemini, Groq or Alibaba Cloud Bailian), and the text it returns replaces Apple's result. We never receive it. With the switch off or no key set, no audio is sent.
On Android, recognition prefers an installed on-device Chinese model. If none is available, it falls back to the standard Android speech recognition service and shows a cloud indicator. You can also prefer that service in Settings. Android documents that it is likely to stream audio to remote servers; the device and speech-service provider control the exact route.
The microphone is on only while you dictate, or while a keyboard voice session is open. The session normally closes after 3 minutes without keyboard activity or dictation; you can explicitly extend this session to 10 minutes or end it at any time. iOS shows its microphone indicator whenever it is on.
We do not keep audio recordings.

The keyboard and “Allow Full Access”

iOS does not let keyboards use the microphone, so the keyboard asks the UTUVO Type app to record for it. They talk through a shared folder on your device (App Group). That on-device link, and haptic feedback, are why the keyboard needs Full Access. Nothing is sent over the network by this link.
The keyboard only inserts your dictation. It does not record or store what you type with other keys or keyboards.
When the microphone starts, iOS shows a back button in the upper-left corner so you can return to the app you were typing in. The keyboard does not read that app's identifier.

Editing and translating

“Say how to change it” and translation use Apple's on-device frameworks (Apple Intelligence, Translation). iOS may download translation language packs from Apple.
Optional smart cleanup sends the recognized transcript and relevant personal-dictionary terms to the provider you configure with your own API key. Field context is off by default; when enabled, a bounded excerpt of text before the cursor is also sent. Android may also send the current app name and your saved tone hint when this option is enabled. On iOS the keyboard cannot provide the host app name. The original recording is not forwarded to the cleanup provider. Your key is stored securely on your device.
On Android, translation uses Google's ML Kit on the device and may download language models. ML Kit processes the text and translation on the device; it sends Google device and app information, per-installation identifiers, performance and usage metrics, error codes, and the configured source and destination languages for diagnostics and usage analytics. Google says these metrics are encrypted in transit and are not shared onward with third parties. Google's ML Kit data disclosure explains these categories. If you configure a separate cloud translation provider with your own API key, the text you translate is sent directly to that provider.

What is stored, and where

Your dictation history, personal dictionary and settings are stored on your device, in UTUVO Type's own storage. They may be included in your iPhone or Mac backups like any app data.
Delete single entries in History (swipe left), or delete everything by deleting the app.

UTUVO Type for Mac

By default transcription and clean-up run locally on your Mac. The speech model is downloaded once, only when you click Install local engine.
If you select Bailian as the speech backend, microphone audio is streamed to the configured Bailian ASR endpoint. A cloud formatter is also available with your own key; it receives the transcript and, only when enabled, bounded app/field context.

Children

UTUVO Type is intended for adults aged 18 and over and is not designed for children.

Contact

Questions or concerns: email support@utuvo.app or open an issue at github.com/mickyyang-1407/utuvo-type/issues. The source code is public, so every statement on this page can be checked.

繁體中文

隱私權政策

最後更新：2026 年 9 月 26 日．適用 UTUVO Type iPhone 版（app 與鍵盤）、Android 版及 Mac 版。

簡單說：UTUVO Type 沒有帳號、廣告或追蹤。我們沒有自行營運分析服務，也沒有接收你聲音或文字的伺服器。Android 使用 Google ML Kit 時，會向 Google 傳送診斷與使用指標。音訊或文字可能直接送到你選用的語音或 AI 服務；路徑與控制方式列在下面。

你的聲音

iPhone 語音使用 Apple Speech，支援時優先在裝置端辨識。可在 設定 → 辨識與隱私 允許 Apple 伺服器辨識；開啟 只用裝置端辨識 會阻止網路辨識，語言不支援時直接拒絕錄音。允許伺服器辨識時，實際路徑由 Apple 決定，適用 Apple 的隱私權政策。聽寫畫面會標示本次請求是否允許雲端。
iPhone 選配雲端辨識（預設關）：你在「設定 → 智慧整理」打開「雲端辨識」後，每次聽寫的錄音會用你自己的 API key 送到你在那裡選的服務（Google Gemini、Groq 或阿里雲百鍊），回傳的文字取代 Apple 的辨識結果。我們收不到這些錄音。沒打開或沒有 key 時，不會送出任何音訊。
Android 預設優先用已安裝的裝置端中文模型。若本機模型不可用，會回退到 Android 標準語音服務並顯示雲端提示；也可在設定裡優先使用該服務。Android 文件指出該服務通常會把音訊串流到遠端，實際路徑由裝置與服務商決定。
只有在你聽寫時、或鍵盤語音工作階段開著時才會開麥克風；沒有鍵盤活動或聽寫時，工作階段通常在 3 分鐘後自動關閉。你可以自行將本次延長到 10 分鐘，或隨時結束。麥克風開著時 iOS 會顯示指示燈。
我們不保留錄音。

鍵盤與「允許完整取用」

iOS 不讓鍵盤使用麥克風，所以鍵盤請 UTUVO Type app 代為錄音，兩者透過你裝置上的共用資料夾（App Group）溝通。鍵盤需要「完整取用」就是為了這條裝置內的連線和觸覺回饋；這條連線不經過網路。
鍵盤只會插入你的聽寫內容，不會記錄或保存你用其他按鍵或鍵盤打的字。
麥克風開好後，iOS 左上角會顯示返回按鈕，讓你回到原本正在打字的 app。鍵盤不會讀取該 app 的識別碼。

改寫與翻譯

翻譯可使用 Apple 的裝置端框架（Translation）；iOS 可能會向 Apple 下載翻譯語言包。
選配智慧整理會把辨識文字和相關個人字典詞條送到你設定的自備 API key 服務。欄位脈絡預設關閉；開啟後會附上游標前受限長度的文字；Android 也可能附上目前 App 名稱和你儲存的語氣提示。iOS 鍵盤無法提供宿主 App 名稱；原始錄音不會轉送到整理服務。key 安全儲存在你的裝置上。
Android 翻譯使用裝置端 Google ML Kit，可能下載語言模型。ML Kit 在裝置端處理原文與譯文；它會向 Google 傳送裝置與 app 資訊、每次安裝的識別碼、效能與使用指標、錯誤碼，以及設定的來源與目標語言，用於診斷和使用分析。Google 表示這些指標傳輸時會加密，且不再分享給第三方；詳見 ML Kit 資料揭露說明。若你另外設定自備 API key 的雲端翻譯服務，翻譯文字會直接送到該服務。

存了什麼、存在哪裡

聽寫歷史、個人字典與設定都存在你的裝置上、UTUVO Type 自己的儲存空間。跟其他 app 資料一樣，可能會被包含在 iPhone 或 Mac 的備份裡。
在「歷史」往左滑可刪單筆；刪除 app 就會全部刪除。

Mac 版

預設轉錄與整理都在你的 Mac 本機執行。語音模型只在你按下 安裝本機引擎 時下載一次。
選取百鍊語音後，麥克風音訊會串流到你設定的百鍊 ASR endpoint。雲端整理使用自備 key，收到逐字稿；只有開啟周邊 context 後才收到受限 App／欄位脈絡。

兒童

UTUVO Type 以 18 歲以上成人為對象，並非為兒童設計。

聯絡

有疑問可寄信至 support@utuvo.app，或在 github.com/mickyyang-1407/utuvo-type/issues 開 issue。程式碼是公開的，這頁的每一句都可以查證。