AI-powered translation, text correction, and prompt engineering — right in your browser.
Install · Features · Usage · Build from source · Privacy
LingFlow AI is an open-source, Manifest V3 browser extension that combines AI translation, text correction, OCR, text-to-speech, and prompt engineering (graphics, video & code) in a single workspace. Work from the popup, side panel, or directly on any page.
v2.0 UI languages: English and Polish (
_locales/en,_locales/pl). Translation targets still include 26 languages in the language picker.
LingFlow AI supports two tiers:
| Tier | What you get | Requirements |
|---|---|---|
| Chrome AI (Free) | Translation via Chrome's built-in Translator API; correction and prompts via Prompt API (limited to supported languages) | Chromium 138+ with Built-in AI. First use downloads the on-device model (~1 GB). No API key. Text stays on-device. |
| BYOK (Bring Your Own Key) | Full translation, correction, tone rewriting, prompt engineering, and OCR via OpenAI or Google Gemini | Your own API key. Text sent to the provider over HTTPS. |
Chrome AI does not send your content to external servers. The browser may need internet on first use to download the on-device model.
LingFlow AI is an open-source project distributed through GitHub only (it is not published in the Chrome Web Store).
- Download the latest
lingflow-ai-v<version>.zipfrom GitHub Releases. - Unzip to a folder.
- Open
chrome://extensions→ Developer mode → Load unpacked → select the folder.
See Build from source.
- Open the extension → Settings.
- Pick your AI engine:
- Chrome AI (Free) — on-device, no key. Model download progress is shown on first use.
- OpenAI or Google Gemini — paste your API key for OCR and full language/tone support.
- Set default target language and save.
API keys: OpenAI · Google Gemini
- 26 target languages with auto-detect and swap.
- Tone & register (BYOK only): Auto, Formal, Casual, Business, Friendly.
- Chrome AI uses the native Translator API; unsupported pairs fall back to Prompt API.
- BYOK providers support all pairs with tone control.
- Dedicated Correct tab: proofreading in the text's own language (auto-detected or chosen).
- Minimal edits: spelling, grammar, punctuation and diacritics are fixed; the author's wording is kept.
- Result shown as an inline diff (removed text in red, added in green) with a change count; Copy gives the clean text.
- Optional tone (formal, casual, business, friendly) for both Chrome AI and BYOK.
- Chrome AI: Prompt API languages only (en, ja, es, de, fr); other languages show how to add a Gemini key.
- Category Pills: Quick switching between Graphics, Video, and Code prompt domains.
- Graphics (Nano Banana & Diffusion models):
- Photo: Structured formula (Subject, Style, Composition, Lighting, Camera, Constraints).
- Graphic design: Vector, 3D icons, minimal, editorial, and branding visual prompts.
- Enhance / Expand: Turn rough text or tag-lists into full descriptive prompts.
- Web UI Mockup: Clean Dribbble/Behance-level UI mockups (strictly no hands/stock lifestyle clutter).
- Mobile App UI: Flat and isometric mobile screen presentations with touch ergonomics.
- 4-Screen UI Collage: Panoramic 16:9 4-screen app showcase (Onboarding, Dashboard, Detail, Profile).
- Video (Veo / Gemini Omni, Sora, Kling, Runway, Luma) - one shot of 5-10 s with subject, action, camera move, light and an audio line (ambient, effects, short quoted dialogue):
- Cinematic: film language for shot size, lens, camera move and color grade.
- Image-to-Video (I2V): motion for an existing frame with subject consistency.
- 3D Product Commercial: studio light, reflections and a single product move.
- Social Media: a hook in the first second, vertical by default.
- Seamless Loop: cyclic motion whose last frame matches the first.
- Code (AI Assistants & Design Directives):
- AI Coding Agent: task brief with Goal, Context, Requirements, Constraints, Edge cases and Done when, in your language.
- UI Aesthetic Spec: design direction with concrete fonts, hex palette, spacing and motion values.
- Image prompts are written as a narrative paragraph, with exclusions phrased positively and without "8k/masterpiece" filler.
- With Chrome AI, ideas in unsupported languages (e.g. Polish) are translated on-device and the prompt is written in English.
- Inline Controls: Direct inline aspect ratio selector (
16:9,9:16,1:1,4:5,21:9) and camera motion selector (zoom-in/out,pan,tracking,orbit,fpv,static).
- Region capture on any page (camera icon or right-click menu).
- OpenAI or Gemini vision models with structured JSON output and high image resolution.
- Transcription + translation with copy and TTS.
- Right-click a selection: Translate with LingFlow or Correct with LingFlow (correction on the page needs a cloud key).
- Result tooltip with copy, listen and in-place replace in editable fields.
- Optional floating button on text selection: Settings → Show translate button when you select text. It asks for access to all sites only when turned on and gives it back when turned off.
- No content script runs by default; LingFlow touches a page only when you invoke it.
- History (max 100 items) with search, filter and pin.
- Settings → Privacy: turn history off, auto-delete entries after 1/7/30/90 days, keep API keys in memory only until the browser closes.
- Web Speech API or Chrome TTS.
- Cut-off answers are flagged; an empty answer that ran out of budget is retried once.
- A retired model switches to the provider default automatically.
- Results are cached for an hour in session memory (also across popup reopenings).
| Task | How |
|---|---|
| Translate | Translate tab → languages → text → action button |
| Fix text | Correct tab → text → Correct |
| Generate prompt | Prompt tab → Category (Graphics/Video/Code) → Type & Parameters → Generate |
| OCR | Camera icon or right-click → Translate text from a screen area → drag region (BYOK) |
| On-page | Select text → right-click → Translate / Correct with LingFlow (or the optional floating button) |
| Side panel | Ctrl+Shift+Y / Cmd+Shift+Y |
- Vanilla ES6+, Tailwind CSS 3.x, Manifest V3
- Chrome Built-in AI,
@google/genai, OpenAI Chat Completions - webpack, Vitest, Playwright
Models and defaults: popup/modules/constants.js
Prerequisites: Node.js 20+ (see .nvmrc), Chromium browser.
git clone https://github.com/dilitS/LinkFlowAI.git
cd LinkFlowAI
npm install
npm run verify # lint + test + build + manifest checks
npm run package # creates releases/lingflow-ai-v<version>.zipLoad unpacked: chrome://extensions → Developer mode → project root folder.
popup/output.cssis committed so the repo loads unpacked without a CSS build; re-runnpm run build:cssafter Tailwind class changes (CI fails if the committed file is stale).
Prompt quality evals (real API, costs a few cents):
GEMINI_API_KEY=... npm run eval -- --judge # or OPENAI_API_KEY=... npm run eval -- --provider openai
npm run eval -- --dry-run # print prompts onlyCases live in evals/cases.json; reports go to evals/results/ (git-ignored).
Manual QA: see SMOKE-TEST.md.
- Chrome AI: on-device processing; model download may use network once.
- BYOK: HTTPS to OpenAI or Google. Keys in
chrome.storage.local(trusted contexts only), or in session memory only if you choose so in Settings. - Minimal permissions: no access to websites by default; the optional all-sites access is only for the floating selection button.
- No telemetry by LingFlow AI.
Full policy: PRIVACY.md · Questions: GitHub issues
- OCR requires BYOK (no on-device vision in v2.0).
- Chrome AI depends on hardware, OS, and browser version.
- Chrome AI Prompt API languages: en, ja, es, de, fr (correction in other languages needs a cloud key).
- Input limit: 5,000 characters with Chrome AI, 30,000 with Gemini/OpenAI.
- UI localized to English and Polish only in v2.0.
- Provider rate limits apply to BYOK.
See ROADMAP.md and PLAN-AKTUALIZACJI-V2.md.
See CONTRIBUTING.md. Run npm run verify before opening a PR.