EdgeTranslate-v3 is a Manifest V3 browser translation extension based on the original Edge Translate project. This fork keeps the familiar selection-translation workflow while modernizing the extension for current Chrome, Firefox, and Safari policies.
The 4.x line focuses on a refreshed Material-inspired UI, a heavily updated pdf.js viewer, dark mode, better translation panel ergonomics, and experimental Chrome on-device AI translation through Gemini Nano where the browser supports it.
- Current repository: Meapri/EdgeTranslate-v3
- Original project: EdgeTranslate/EdgeTranslate
- Latest release: v4.1
- Selection translation in a side panel with copy, edit, pin, resize, and TTS actions.
- Word assistance with translation, detailed meanings, definitions, and examples when the selected provider supports them.
- Chrome built-in AI translation support for regular text and word translation flows.
- Google and other practical engines remain available for full-page translation.
- Built-in pdf.js viewer with selection translation inside PDFs.
- Material-inspired PDF toolbar, menus, dialogs, page sidebar, and dark mode.
- Dark mode support across the PDF viewer, translation panel, popup, and settings page.
- Settings page cleanup with clearer provider names and less visual clutter.
- Keyboard shortcuts, blacklist controls, and configurable translation button behavior.
Chrome's on-device Gemini Nano path is exposed as an AI translation provider where the
browser makes the LanguageModel API available. It is useful for smaller selections and
word-assistance tasks, but it is not used as a full-page translation engine in this fork.
Google AI Studio (Gemini 2.5 Flash Lite) is also available as a server-side AI translation provider with streamlined prompts optimized for minimal token usage.
On-device pipeline (v4.1):
- Single-pass streaming translation — no redundant refinement passes.
- Smart text chunking splits long inputs at paragraph, line, and sentence boundaries (800-character optimal window) to maintain quality and formatting.
- System prompt enforces proper noun preservation and prevents literal Kanji→Hanja hallucinations.
- Post-translation rules catch known false-friend errors as a final safety net.
- Passthrough punctuation system preserves decorative symbols (※, →, ★, etc.) that models tend to drop during translation.
AI Studio pipeline (v4.1):
- Concise prompts tuned for Gemini 2.5 Flash Lite with thinking disabled
(
thinkingBudget: 0) for maximum speed. - Script constraint rules ensure output uses the correct writing system.
- Dictionary mode returns structured JSON with meanings, definitions, and examples for single-word queries.
Chrome's LanguageModel API and Gemini Nano availability depend on the user's Chrome
version, device, feature flags, and model download state. When unavailable, the extension
falls back to the configured provider path rather than treating AI translation as a required
runtime dependency.
EdgeTranslate-v3 opens supported PDF links in its bundled pdf.js viewer so text selection
and translation work inside documents. The 4.x viewer uses pdf.js 5.7.284 and includes
many layout and interaction fixes.
PDF behavior:
- Web PDF links can be redirected into the extension viewer for translation.
- Local PDFs can be opened when Chrome's "Allow access to file URLs" setting is enabled for the extension.
- Dragged PDF files and blob-backed viewer URLs are preloaded or repaired so pdf.js can load them reliably.
- The viewer includes a native-viewer bypass action for users who want to open the current document outside EdgeTranslate.
- The PDF detection fallback avoids probing unrelated non-PDF pages, preventing noisy CORS errors on pages such as Chrome Web Store developer console URLs.
- Selection translation
- PDF viewer and PDF selection translation
- Google full-page translation
- Chrome built-in AI translation when the browser/device supports it
- Offscreen document support for extension-context AI translation
- Selection translation
- PDF viewer and PDF selection translation with browser-specific limitations
- No Chrome built-in AI translation
- No Chrome-only full-page translation behavior
- Selection translation and PDF viewer support through the Safari extension build path
- No Chrome built-in AI translation
- Safari release requires the Xcode project and Apple credentials
Release assets usually include:
edge_translate_chrome.zipedge_translate_firefox.xpi
Use the repository root as the working directory.
npm ciRun the unit test suite:
npm testRun the EdgeTranslate workspace tests directly:
npm test -w edge_translate -- --runInBandBuild the default Chrome target:
npm run build:chromeBuild individual targets:
npm run build:chrome
npm run build:firefox
npm run build:safariCreate browser packages:
npm run pack:chrome -w edge_translate
npm run pack:firefox -w edge_translateValidate the Firefox package:
npm run lint:firefoxBuild outputs:
- Chrome unpacked build:
packages/EdgeTranslate/build/chrome/ - Firefox unpacked build:
packages/EdgeTranslate/build/firefox/ - Chrome package:
packages/EdgeTranslate/build/edge_translate_chrome.zip - Firefox package:
packages/EdgeTranslate/build/edge_translate_firefox.xpi - Safari build output:
packages/EdgeTranslate/build/safari/
- Open
chrome://extensions. - Enable Developer mode.
- Click "Load unpacked".
- Select
packages/EdgeTranslate/build/chrome/. - Enable "Allow access to file URLs" if you want local PDF/file translation.
- Open
about:debugging. - Choose "This Firefox".
- Click "Load Temporary Add-on".
- Select any file inside
packages/EdgeTranslate/build/firefox/, or use the generatededge_translate_firefox.xpi.
Safari builds live under packages/EdgeTranslate/safari-xcode/.
Useful commands:
npm run build:safari
npm run safari:sync -w edge_translate
npm run safari:release -w edge_translatesafari:release builds, syncs resources into the Xcode project, archives, exports, and
uploads. It requires valid App Store credentials.
The extension uses host access so selection translation, PDF detection, and content scripts
can work on user pages. Chrome builds also request offscreen so Gemini Nano translation can
run from an extension document context when page injection is not available, such as inside
the extension PDF viewer.
Main permission categories:
activeTab,tabs, andscriptingfor selection translation and page interactions.contextMenusandstoragefor commands and user settings.webNavigationandwebRequestfor PDF detection and viewer routing.offscreenon Chrome for extension-context AI translation.
- No analytics or telemetry collection is added by this fork.
- Translation text is sent only to the provider selected/configured by the user.
- Chrome on-device AI translation runs through Chrome's built-in local model path when available.
- File URL access is optional and controlled by the browser extension settings.
For a normal release:
- Update
packages/EdgeTranslate/package.json. - Update
package-lock.json. - Update
packages/EdgeTranslate/src/manifest.json. - Run tests and browser packaging.
- Create a release commit and tag, for example
v4.0.1. - Upload Chrome and Firefox package artifacts to GitHub Releases.
See RELEASE_CHECKLIST.md for the older project checklist.
Legacy feature references from the original project are still useful for general behavior:
This project follows the same license structure as the original Edge Translate project.
Thanks to the original Edge Translate maintainers and contributors. EdgeTranslate-v3 keeps that foundation alive for Manifest V3 browsers while continuing to adapt the extension for modern browser APIs, PDF workflows, and AI-assisted translation.