Read in two languages
Stream translated text beneath its original paragraph. Nearby passages are processed first, and recent results can be reused without another API request.
For Microsoft Edge & Google Chrome
Read beyond
language.
A quieter way to translate the web. Keep the original in view, choose your translation provider, and read at your own pace.
Your API or a limited key-free basic trial
Current extension interface: Simplified Chinese
YOUR WORDS. YOUR PROVIDER.QF / 01
One workflow. Your choice of engine.
Google CloudDeepLMicrosoft TranslatorOpenAI-compatibleCustom JSON
Your connection, directly
Configure your endpoint and credentials. Translation requests go from the extension to your chosen provider, with permission for that API domain.
Select, then translate
Enable the selection button on a site, select a phrase, and click Translate. Read the result beside your selection. No request is sent just by selecting text.
VERSION 0.8.16
Expand the evidence. Read the full answer.
Paper and PDF/Word/TXT/Markdown workbenches now expand into a dedicated view. Larger figure previews and a sticky search/bulk toolbar keep selection convenient. Questions and answers use separate columns; narrow screens stack them. Escape returns to reading while retaining selections, drafts, results and active analyses. AI candidate review provides larger previews and source-page access.
Your colors. Your reading shortcut.
Choose immersive text colors with light/dark previews. The webpage shortcut defaults to Shift+A: type another combination and save, or record it. Reset or disable it at any time. Inputs and editors keep normal typing. These preferences apply without retranslating or changing API profiles.
Read the content in front of you.
Immersive translation prioritizes the live viewport and supplements scrolling, expanded and dynamically changed text. Confidently detected target-language text is skipped. Use the nearby opt-in selection button for a passage. Browser-internal pages and built-in PDF viewers use the separate extension entry points.
Keep the page. Translate in place.
Open PDF, Word DOCX, TXT or Markdown locally or from a permitted URL. Choose in-place immersion, original-layout comparison or paired paragraphs. PDF page proportions, images and charts stay in position; long or unsupported text has a full-translation fallback. Choose page-by-page or continuous PDF reading. Translate the current page or all pages and export bilingual TXT.
Files are limited to 25 MB; PDF/rendered Word retain a 300-page bound. Extracted body text and TXT/Markdown have no fixed character cutoff, but provider context and browser memory limits remain. Word layouts can differ from Microsoft Word. Scanned body replacement needs external OCR. Layout-preserving PDF/DOCX export is not included.
Ask about papers and their evidence.
Use a movable, resizable paper Q&A window or the document research workflow. Caption/number-aware extraction inventories figures, image tables and native/ARIA tables. Search or select all figures, batch-load available high-resolution originals, and combine selected figures with the paper text. Native tables copy locally as Markdown or TSV. Image tables require explicit vision-AI transcription and checking. For PDF, extract current/all pages using captions and drawing geometry; AI supplementation lives in the figure overview and requires candidate review. Local/AI results share crop/table deduplication. For missing material, import/paste a screenshot or crop manually.
Capture, selection and loading do not call AI. Optional analysis explains figures and creates downloadable Python/R drafts; code is never executed. Editable Skill prompts and project-isolated cross-page notebook memory are available; memory is off by default and exportable in Settings. Automatic extraction is heuristic and cannot guarantee every website. Up to four original images can accompany a paper question; loading/capture resource bounds remain.
Your provider. An informed model choice.
Save up to 20 API profiles for Google, DeepL, Microsoft, OpenAI-compatible or custom JSON services. Fold settings sections individually. Create/switch API connections only in Settings. Reading views discover/refresh the active API’s model list and switch the shared model using plain-ID dropdowns. Test a draft before saving; names alone do not establish capabilities. MyMemory offers a limited key-free trial, typically 5,000 anonymous characters/day; it is not unlimited and may retain submitted text.
Translation, model discovery, tests and AI analysis have no extension deadline. Actual HTTP/network/stream errors or Stop end a call; provider limits still apply. Separate document/image downloads retain safeguards. Site/API access is requested as needed; an explicit optional all-websites grant reduces prompts.
Verified behavior, clear limits.
0.8.16 passed 87 behavior checks plus package/syntax validation, using isolated Edge, generated papers and synthetic APIs. The checks cover expanded layout/state, PDF candidate review, web analysis and supplementation. Earlier 0.8.15 passed 121 checks: 110 behavior checks and 11 build/package checks, with generated documents, saved fixtures and synthetic local APIs. Live publisher availability, real-provider speed/accuracy and current Chrome integration are not established. The bare-ID follow-up reran 27 model checks successfully without adding to the total. Earlier PDF/full-suite results are recorded separately. Files parse locally; only requested text or chosen images go to your configured provider. No developer translation proxy, ads or analytics. Verification →
Community & support
Join the conversation.
Share feedback, ask questions and discuss QudadFlow Translator.
QudadFlow Translator is independent and not affiliated with the listed providers. Provider accounts, fees and privacy terms apply. GitHub release and store review states are tracked separately in the 0.8.16 submission record.