FlowRead 1.0Download original Markdown

FlowRead Privacy Policy

Effective: 19 November 2026. Product: FlowRead 1.0 for Microsoft Edge. Maintainer and privacy/support contact: admin@4dv.top. Source. FlowRead is an independent GPL fork of Read Frog, not an official upstream release. This policy describes this fork.

Data on your device

Settings, your OpenAI API key, translation caches, local document history, reading positions, PDF annotations/bookmarks, vocabulary notebooks, learning answers/statistics and optional saved reading recordings are stored in your browser's local storage/IndexedDB. The key is kept in separate local credential storage, excluded from normal settings exports. Local browser storage is not a hardware vault; people with access to your profile may access it.

Docs opens local PDFs in bundled PDFCraft Web. PDF.js reads text and renders images locally for FlowRead's dictionary, page summaries and page-range image translation. The PDF binary stays on your device. Explicit AI actions send selected text, a bounded page excerpt or the chosen page images directly to OpenAI. Reading positions and translation history are stored in the browser profile. Embedded PDF JavaScript is disabled in the FlowRead build of PDFCraft. PDFCraft edits stay in the current viewer session. Bundled OCR remains available for local text measurement in Image Translation.

Docs opens DOCX in a bundled copy of WordCraft Web. The viewer processes the document and its comments locally in tab memory. Edits there are temporary and the original file stays unchanged. Display and reading preferences, such as theme, zoom and navigation, are kept in the current browser profile. Explicit actions can send selected document or comment text to OpenAI for translation, technical explanation, summary or listening. Page or visible-area captures are prepared locally and opened in the crop editor. The image enters an OpenAI request only when you explicitly request image recognition or translation. Completed translations and source excerpts can be saved, exported or cleared in the local Word translation history. PDF and WordCraft Web can save local source positions for returning to highlighted excerpts. The DOCX file, path, file name and comment author metadata are excluded from provider requests.

Word study notes, highlights and bookmarks store their text excerpts and anchors in local extension storage, keyed by the document hash. They remain available when reopening the same document in the same browser profile and can be deleted from Word Reader. The most recent reading positions and zoom levels are stored locally. They are bounded to 500 documents.

Docs uses bundled GridCraft Web to open XLSX workbooks, switch sheets and search offline. A bundled basic reader is available when GridCraft cannot open a workbook. The workbook is processed locally and temporary edits do not overwrite the file. Viewing preferences are stored locally. When you explicitly ask for a translation, FlowRead sends the displayed values in the selected cell range or the chosen cell's note text to OpenAI. The workbook, file name, note author and note replies are excluded from the request. Completed translations, source excerpts and cell locations are stored in local translation history. Grid captures open locally in the shared crop editor. Only an explicit image recognition or translation action sends the crop to OpenAI. Docs shows recently opened PDF, DOCX and XLSX files together. The list is local and keeps file names, sizes, content hashes, open times, and a browser file handle when the local picker grants one. Removing a list entry leaves the original file on your device.

Local diagnostics keep a bounded log of extension events (up to 3,000 records), including site hosts, error metadata and short text previews; full web URLs are reduced to origins, file URLs lose paths, and known secret fields, request headers and request/response bodies are scrubbed. The in-page diagnosis view reads paragraph verdicts locally and does not translate or upload content. An explicit diagnostics export contains sanitized logs, public settings and local table counts. Nothing is sent automatically: review the exported JSON before sharing it, since settings, snippets or sites can still be personal. These logs are separate from learning/usage histories and browser-account sync remains removed.

Direct OpenAI processing

You supply your own OpenAI API key. AI translation, dictionary/enrichment, summary, speech generation, image OCR, pronunciation grading, audio transcription and video analysis send required text, selected image crops or audio directly to fixed OpenAI HTTPS/WSS endpoints, authenticated with that key. There is no FlowRead content proxy, developer processing server, analytics SDK, advertising or data sale.

Text may include selected passages, neighboring page/caption context, glossary terms or your optional instructions. Video analysis may send a bounded transcript excerpt/title plus enabled audio samples for context and speaker slots. Unknown speakers use a configured preset voice. This is not voice cloning or identity recognition. Selected content can contain personal or sensitive information; share only material you are authorized to send.

Only crops explicitly chosen for OpenAI image/OCR processing are sent. Local OCR remains local. File transcription sends the chosen audio; live tab transcription sends audio while capture is active. Pronunciation microphone audio uploads only after explicit grading. Local shadowing recordings are not uploaded for scoring. Tab capture does not use the microphone. Microphone practice asks for browser permission and has separate recording/grading controls.

The explicit OpenAI translated image mode sends the selected crop directly to OpenAI Images Edits, which generates an edited PNG. The extension displays/downloads that PNG unchanged; this mode does not run local OCR or local text replacement. The translation reuse cache stays in bounded tab memory for up to 30 minutes. Completed PNGs can be saved separately in local image history. Regenerating is a separate paid action; cancellation cannot retract an already received/billed request. Image generation does not use the text-mode glossary/custom prompts and may change image details despite layout-preservation instructions.

Translated image PDF is a separate explicit page-range action. The local reader renders full-page PNG images, then sends those selected page images sequentially to OpenAI Images Edits. The original PDF binary, filename/path and FlowRead sidecar annotations are not sent. A full-page image may include text excluded from paragraph translation; this mode is not redaction and the UI discloses the full-page scope before starting. Image reuse stays in bounded document-tab memory (up to 20 images/96 MiB encoded characters) until cleared, configuration/document changes or tab close. Viewing/downloading PNGs and locally assembling a PDF/ZIP do not call OpenAI. Each requested image generation has separate API fees.

Local image history saves only finished translated PNGs from these two modes, dimensions, timestamps, and synthetic page groups and numbers for PDF results. By default new images are kept for seven days. You can choose one or 30 days, turn off future saves, or delete history from the history page. The oldest images may be removed sooner when the limit of 80 images or 256 MiB is reached. Original images, PDFs, filenames, paths, instructions, and credentials are outside this history. Expired images are unavailable and removed by local cleanup. Viewing and downloading history stay on the device.

AI features have OpenAI usage fees, separate from ChatGPT subscriptions. They require a configured key and user initiation/enablement. User-enabled automatic page/site translation can send eligible text on matching pages. Live transcription and automatic video analysis continue within their owned session until stopped or invalidated. Automatic video analysis samples the user-selected amount (15 seconds to ten minutes, default 15 seconds) of played audio in each new ten-minute timeline part, at most 128 parts/session. Explicit prepared mode records a selected interval from the current position, waits for analysis, then restores the owned position paused. Longer intervals take more time and may cost more. Changing duration stops old work and turns AI off until re-enabled. These modes do not download future audio or silently fall back to paid OCR/ASR. A failed paid analysis does not automatically retry. Turn the mode off to stop it.

OpenAI handles received data under its own terms and account/endpoint retention settings: OpenAI data controls. store:false where supported does not guarantee zero retention. FlowRead cannot control provider retention; review provider terms before enabling AI.

Other network access

Optional PDF image-edit instructions are sent directly to OpenAI with each selected page only when you start generation. They stay in the current document tab's memory, outside settings, sync, exports and usage logs. Local cost estimates read page geometry and use a bundled pricing snapshot; they do not upload content or call a pricing service. Estimates cover image output only and exclude unknown image/text input fees, taxes and future price changes; they are not actual billed usage.

Native caption requests go to the watched website and its caption hosts. Opening a remote PDF requires a source GET to its URL; this is not uploading the document. Site access enables translation interfaces on chosen sites and supported local files when file access is granted. Scripts, WASM, OCR models, fonts, icons and site rules ship in the package, without remote rule updates or executable CDN fallbacks.

Browser-account configuration sync has been removed. FlowRead no longer reads or writes browser storage.sync or runs recurring configuration-sync jobs. Settings export/import and backups are local files/local browser data, not cloud sync. Previously stored sync data may remain in your browser/account; this update does not access or erase it. FlowRead has no independent cloud account. Support/source links open external websites. Emailing support or posting an issue shares what you send with those services/the maintainer; remove keys and private content first.

Retention and controls

Session-only captures, live transcripts and video-analysis profiles are memory-only and released when their owner/session ends. Save/download controls create local files when requested. Persisted local data remains until feature-specific clearing, browser data clearing or uninstall, subject to cache limits. Closing a tab does not delete saved notebooks/history. Uninstall normally removes extension browser data, not exported files. Deleting a key stops future authenticated requests but cannot retract data already sent. This version does not provide controls to access or erase previously synchronized settings. Browser/account data controls are separate. Revoke site, microphone or file access in browser settings.

There is no developer-held content account to erase. Privacy questions: admin@4dv.top. FlowRead is not directed to children; the maintainer does not knowingly solicit children's information. Data use is limited to the disclosed user-facing functions: no personalized advertising, sale, creditworthiness or lending use. This policy must be updated before adding new data uses.

Tóm tắt tiếng Việt

FlowRead dùng khóa OpenAI của bạn; thao tác AI gửi văn bản/ngữ cảnh, vùng ảnh hoặc audio cần thiết trực tiếp đến OpenAI và có phí API riêng. PDF gốc xử lý trên máy, không upload cả PDF. OCR cục bộ, tìm kiếm và chú thích không gọi API. Chế độ tự động chỉ chạy sau khi bạn bật. Sổ từ, lịch sử, vị trí đọc và khóa lưu trên máy; đã gỡ đồng bộ cấu hình qua tài khoản Edge; xuất/nhập file và sao lưu trên máy vẫn còn. Bản này không truy cập hoặc xóa dữ liệu đồng bộ cũ. Không có máy chủ xử lý nội dung của FlowRead, quảng cáo hay telemetry. Xóa dữ liệu bằng nút của tính năng hoặc cài đặt trình duyệt; liên hệ admin@4dv.top. Lưu dữ liệu phía OpenAI phụ thuộc chính sách/tài khoản/endpoint, không cam kết lưu trữ bằng không.