英語救星 隱私權政策
適用平台：iOS、iPadOS 與 visionOS
最後更新：2026 年 10 月 14 日
開發者：Yu-Hsiang Chang
英語救星提供英文練習、字典解釋、故事閱讀與插圖功能。App 不要求註冊開發者提供的帳號。以下說明 iOS 1.42 與 visionOS 1.3 版各項功能的資料去向、用途與可用的控制方式；可用功能依 App 與系統版本而異。
1. 裝置端文字功能
預設文字模式使用 Foundation Models 框架的裝置端模型，產生練習建議、字典解釋、故事及插圖描述摘要。這條文字推論路徑在裝置上執行，需支援的裝置、語言、系統設定與已下載的模型。若無法使用，不會自行將請求轉送至外部文字服務。
2. 系統故事插圖
按「產生插圖」後，App 透過 Image Playground 框架開啟系統介面，傳入由故事整理出的簡短圖像描述。你可以修改描述、選擇圖片或取消。在 iOS 27 與 visionOS 27，生圖使用 Apple 提供的 Private Cloud Compute，需要網路連線；較早版本依該系統提供的方式處理。此流程不經開發者自建的生圖伺服器或中繼服務。
App 不啟用外部服務商的生圖風格與人物個人化功能，也不提供來源照片。取消不會刪除故事或既有插圖。接受圖片後，App 會先複製暫存圖片；收藏故事時，再把文字與圖片存至 App 本地儲存空間。
3. 自備金鑰的外部文字服務（BYOK）
你可以自備 API 金鑰，選擇 Google Gemini API 或 Groq。App 會在啟用前顯示資料說明，並向各服務商分別取得資料傳送同意。未同意時，不會向該服務商驗證金鑰或送出文字生成請求。
請求由裝置直接透過 HTTPS 傳送到選定服務商：Google 的 generativelanguage.googleapis.com 或 Groq 的 api.groq.com，不經開發者自建伺服器。金鑰保存在裝置 Keychain，並提供給對應服務商驗證請求。
各功能送出的內容如下：
• 字典：所查單字、可用的句子上下文，用於產生解釋與例句。
• 練習：你想表達的意思與輸入的英文，用於產生回饋及詞彙內容。
• 故事：最近學過的單字與選定主題，用於生成故事及理解題。
• 各請求也包含語言偏好及任務指令。
• 金鑰驗證：金鑰及固定測試請求或模型查詢，不包含學習紀錄。
這些 BYOK 文字請求不包含原始錄音或圖片。若你在文字中輸入個人資料，該文字會依你選定並同意的服務路徑處理。
服務商的資料保留、模型改進、服務額度與費用依其政策及你的帳戶設定而定。BYOK 不表示資料留在裝置端，也不適用系統生圖的 Private Cloud Compute 保護。
Google Gemini API：Gemini API Terms
Groq：Groq: Your Data
4. 同意與控制
你可以在 AI 設定中關閉個別服務商的資料傳送同意、改用裝置端文字模式，或移除 API 金鑰。關閉同意會阻止新的請求，但無法撤回已完成的傳送，也不會刪除服務商已收到的資料；相關資料與帳戶管理請依服務商提供的方式處理。原有金鑰會在升級時保留，但舊版的概括風險同意不會自動轉成新版資料傳送同意。
5. 語音功能
iOS 的選用語音輸入會在需要時請求麥克風及語音辨識權限，透過系統 Speech 框架辨識。依裝置支援及設定，辨識可能使用系統語音服務。辨識後的文字依你選定的文字模式處理。朗讀使用系統文字轉語音功能。裝置端文字模式不代表所有語音與圖像功能都能離線運作。
6. 本地保存與刪除
練習紀錄、收藏故事及插圖保存在 App 本地容器，可於 App 內刪除。移除 App 通常也會移除本地資料；系統備份可能保留相關檔案，依你的系統設定處理。移除 App 不等於撤銷服務商金鑰或刪除已傳送至服務商的資料。
7. 聯絡與政策更新
如對本政策或資料處理有疑問，請來信 mibsteven.chang@gmail.com 聯絡開發者。政策如因功能調整而更新，會同步調整本頁更新日期。
文中的框架與服務名稱用於辨識處理技術及資料接收方，不表示贊助、背書或合作關係。
EnglishSaver Privacy Policy
Platforms: iOS, iPadOS and visionOS
Last updated: October 14, 2026
Developer: Yu-Hsiang Chang
EnglishSaver provides English practice, dictionary explanations, stories and illustrations. It does not require an account operated by the developer. This policy explains the processing paths and controls for iOS 1.42 and visionOS 1.3; available features depend on your app and system version.
1. On-device text
The default text mode uses the Foundation Models framework's on-device model for practice feedback, dictionary explanations, stories and illustration briefs. This text inference runs on your device and requires a supported device, language, system settings and downloaded models. An unavailable on-device request is not automatically sent to an external text provider.
2. System story illustrations
When you choose Create Illustration, the app opens the system interface through the Image Playground framework with a short image description derived from the story. You can edit the description, accept an image or cancel. On iOS 27 and visionOS 27, image generation uses Apple's Private Cloud Compute and requires internet access. Earlier supported versions use their available system image service. The developer operates no image server or relay.
The app does not enable external-provider image styles or person personalization and supplies no source photo. Cancelling preserves your story and any existing illustration. Accepted results are copied from the system's temporary location; collecting a story saves its text and image in the app's local storage.
3. Optional external text providers (BYOK)
You can supply your own API key for Google Gemini API or Groq. The app explains the data and purpose and requires separate consent for each provider before key validation or text generation requests.
Requests go directly over HTTPS from the device to generativelanguage.googleapis.com or api.groq.com, without a developer-operated backend. Keys are kept in the device's Keychain and sent to the corresponding provider to authenticate requests.
Request contents and purposes:
• Dictionary: the selected word and available sentence context, for explanations and examples.
• Practice: your intended meaning and English attempt, for feedback and vocabulary.
• Story: recent vocabulary and the selected theme, for stories and comprehension questions.
• Requests also include language preferences and task instructions.
• Key validation: authentication and a fixed test request or model check, without learning history.
These BYOK text clients do not send raw microphone audio or images. If you include personal information in your text, that text follows the provider path you selected and consented to.
Retention, model improvement, quotas and charges depend on the provider's policies and your account settings. BYOK does not mean data stays on your device and is not covered by the Private Cloud Compute protections of system image generation.
Google Gemini API: Gemini API Terms
Groq: Groq: Your Data
4. Consent and controls
In AI Settings you can revoke consent for an individual provider, select on-device text mode or remove an API key. Revocation stops new requests but cannot retract completed transmissions or delete data a provider has already received. Use that provider's data and account controls for those requests. Existing keys are preserved on upgrade, but the former generic risk disclaimer is not reused as consent to the new disclosure.
5. Speech
Optional iOS dictation requests microphone and speech-recognition permission when needed and uses the system Speech framework. Depending on device support and settings, recognition may use the system speech service. Recognized text follows your selected text-provider path. Read-aloud uses system text-to-speech. On-device text inference does not mean every speech and image feature works offline.
6. Local storage and deletion
Practice records, collected stories and images are stored in the app's local container and can be deleted in the app. Removing the app normally removes its local files; system backups may retain them according to your settings. Removing the app does not revoke provider keys or delete data previously sent to providers.
7. Contact and changes
For privacy questions, contact mibsteven.chang@gmail.com. We update the date on this page when this policy changes.
Framework and service names identify the processing technology and recipients. They do not imply sponsorship, endorsement or partnership.