Hybrid AI cognitive extension that turns browsing history into semantic memory

A Chrome extension built for the Google Chrome Built-in AI Challenge 2025 that uses a hybrid AI strategy: Chrome Built-in AI (Gemini Nano) when available, Gemini API as reliable cloud fallback. This creates an intelligent, searchable memory of your browsing history with production-ready reliability.

What makes this extension special:

• 🎯 Production Ready: Works reliably TODAY, even in Chrome builds without AI APIs
• 🔄 Automatic Fallback: Tries Chrome Built-in AI first, seamlessly falls back to Gemini API
• 🔒 Privacy First: On-device processing when available, secure cloud fallback when needed
• 🚀 Future Proof: Gets better as Chrome AI rolls out, never breaks
• ✅ All 6 APIs: Complete implementation of all Chrome Built-in AI capabilities

• 🤖 Hybrid AI: Chrome Built-in AI (primary) + Gemini API (fallback) for 100% reliability
• 🧠 Semantic Memory: Convert web pages into searchable semantic memories with AI embeddings
• 📝 AI Summarization: Automatic 2-3 sentence summaries of every page you visit
• 🌐 AI Translation: Translate memories to any language
• ✏️ AI Rewriting: Reformat summaries with different tones, lengths, and formats
• 🔍 Natural Language Search: Query your history using plain English via omnibox (lw <query>)
• 🔐 AES-GCM Encryption: Summaries encrypted locally using Web Crypto API
• ⚡ Vector Similarity Search: Find relevant pages using cosine similarity
• 🎨 Beautiful UI: Clean, modern interface built with Svelte and TailwindCSS
• 📊 Statistics Dashboard: Track your knowledge base growth

• Architecture
• Tech Stack
• Installation
• Usage
• Project Structure
• Development
• Chrome AI APIs
• Privacy & Security
• Contributing
• License

┌─────────────────────────────────────────────────────────────┐
│                    Chrome Extension                         │
│                                                             │
│  ┌───────────────┐   ┌─────────────┐   ┌────────────────┐  │
│  │ Content       │   │ AI          │   │ Vector         │  │
│  │ Scraper       │───│ Summarizer  │───│ Search Engine  │  │
│  └───────────────┘   └─────────────┘   └────────────────┘  │
│          │                  │                    │          │
│          ▼                  ▼                    ▼          │
│     Page Text         Summaries +           IndexedDB       │
│                       Embeddings            (Encrypted)     │
│          │                  │                    │          │
│          └─────────► Query Handler ◄─────────────┘          │
│                          │                                  │
│                     Omnibox & Popup                         │
└─────────────────────────────────────────────────────────────┘

1. Capture: When you visit a page, content script extracts main text
2. Summarize: Background worker uses Chrome AI Summarizer to compress content
3. Embed: Text is converted to vector embeddings using Chrome AI
4. Encrypt: Summary is encrypted with AES-GCM
5. Store: Memory is saved to IndexedDB with embedding
6. Query: Use omnibox to search memories semantically
7. Answer: AI Writer composes answer from relevant memories

• TypeScript: Type-safe development
• Vite: Lightning-fast build tool
• Svelte: Reactive UI framework
• TailwindCSS: Utility-first styling

• chrome.ai.summarizer - Compress page content
• chrome.ai.prompt - Generate embeddings
• chrome.ai.writer - Compose answers
• chrome.ai.translator - Multilingual support
• chrome.ai.proofreader - Clean responses
• chrome.tabs - Page monitoring
• chrome.scripting - Content injection
• chrome.omnibox - Search interface
• chrome.storage - Configuration sync

• Dexie.js: IndexedDB abstraction
• Web Crypto API: AES-GCM encryption
• Vector Search: Cosine similarity

• ESLint + Prettier: Code quality
• Vitest: Unit testing
• web-ext: Development workflow

The extension uses a hybrid AI strategy and works right away:

1. Install the extension
2. Configure Gemini API key (free from Google AI Studio)
3. Start browsing!

Chrome Built-in AI is optional but recommended for privacy.

• Chrome Canary (v143+) or Chrome Stable (when AI APIs launch)
• Node.js (v18+)
• Gemini API Key (free tier available)

No Chrome AI setup needed! This is the fastest way to get started:

1. Build extension (see below)
2. Load in Chrome
3. Open Settings tab
4. Get free API key: Google AI Studio
5. Paste key and save
6. Start browsing!

✅ Works in: Chrome Stable, Chrome Canary, any Chrome 120+
✅ Setup time: 5 minutes
✅ Reliability: 100% (cloud-based)

For maximum privacy and performance:

1. Enable Chrome AI APIs (detailed guide)

   • Navigate to chrome://flags/#optimization-guide-on-device-model
   • Set to "Enabled BypassPerfRequirement"
   • Navigate to chrome://flags/#prompt-api-for-gemini-nano
   • Set to "Enabled"
   • Restart Chrome

2. Build & configure extension

3. Extension will automatically use Chrome AI when available

4. Falls back to Gemini API seamlessly

✅ Works in: Chrome Canary (143+)
✅ Privacy: On-device processing
✅ Performance: Instant responses
✅ Fallback: Gemini API if Chrome AI unavailable

# Clone repository
git clone https://github.com/Sukhtumur/neurochrome.git
cd neurochrome

# Install dependencies
npm install

# Build for production
npm run build

1. Open chrome://extensions/
2. Enable "Developer mode"
3. Click "Load unpacked"
4. Select the dist folder
5. Click extension icon → Settings tab
6. Configure Gemini API key

Open Settings tab in extension popup:

• Provider Badge: Shows "Using Chrome AI" (green) or "Using Gemini API" (purple)
• All 6 APIs: Should show green checkmarks ✓
• Status Message: Confirms active AI provider

See TESTING_GUIDE.md for detailed testing steps.

By default, the extension automatically captures and indexes pages you visit:

1. Browse normally
2. Extension extracts content in background
3. AI generates summary and embedding
4. Memory stored locally (encrypted)

Type brain in Chrome's address bar, then your query:

brain what did I read about machine learning yesterday?
brain show me articles about climate change
brain find that recipe I saw last week

Click the extension icon to:

• Search: Manual query with detailed results
• Memories: Browse and manage stored pages
• Statistics: View storage stats and insights

local-web-brain/
├── src/
│   ├── background/           # Service worker
│   │   └── background.ts
│   ├── content/              # Content scripts
│   │   └── content-script.ts
│   ├── popup/                # Popup UI
│   │   ├── App.svelte
│   │   ├── popup.ts
│   │   └── index.html
│   ├── components/           # Svelte components
│   │   ├── SearchBar.svelte
│   │   ├── MemoryList.svelte
│   │   └── Stats.svelte
│   ├── lib/                  # Core business logic
│   │   ├── ai/              # Chrome AI services
│   │   │   └── chrome-ai.ts
│   │   ├── db/              # Database layer
│   │   │   ├── database.ts
│   │   │   └── repository.ts
│   │   ├── vector/          # Vector search
│   │   │   └── index.ts
│   │   ├── crypto/          # Encryption
│   │   │   └── index.ts
│   │   └── query/           # Query orchestration
│   │       └── index.ts
│   ├── types/               # TypeScript definitions
│   │   ├── index.ts
│   │   └── chrome-ai.d.ts
│   ├── utils/               # Utilities
│   │   ├── helpers.ts
│   │   └── logger.ts
│   └── tests/               # Unit tests
│       ├── setup.ts
│       └── helpers.test.ts
├── public/
│   ├── manifest.json
│   └── icons/
├── vite.config.ts
├── tsconfig.json
├── tailwind.config.js
└── package.json

1. Presentation (components/, popup/): UI components
2. Application (lib/query/): Business logic orchestration
3. Domain (types/): Core entities and interfaces
4. Infrastructure (lib/ai/, lib/db/, lib/crypto/): External services
5. Utilities (utils/): Shared helpers

# Development with hot reload
npm run dev

# Production build
npm run build

# Run tests
npm test

# Lint code
npm run lint

# Format code
npm run format

# Test with web-ext
npm run web-ext

• Clean Code: Descriptive names, single responsibility
• SOLID Principles: Dependency injection, interface segregation
• Type Safety: Strict TypeScript, no any
• Error Handling: Custom error types, try-catch blocks
• Documentation: JSDoc comments, inline explanations

# Run all tests
npm test

# Watch mode
npm test -- --watch

# Coverage
npm test -- --coverage

The extension uses an intelligent hybrid approach:

// Automatically tries Chrome AI first, falls back to Gemini
const summary = await hybridAI.summarize(text)
const embedding = await hybridAI.embed(text)
const translated = await hybridAI.translate(text, 'es')

✅ Summarizer: Compress page content to 2-3 sentences
✅ Embeddings: Convert text to semantic vectors (384-dim)
✅ Writer: Generate natural language responses
✅ Translator: Translate to any language
✅ Proofreader: Clean and correct text
✅ Rewriter: Reformat with tone/length/format options

Chrome AI Available → Use on-device processing (instant, private)
Chrome AI Unavailable → Use Gemini API (1-2s delay, reliable)
Both Unavailable → Show helpful error message

// Initialize hybrid service
await hybridAI.initialize({
  apiKey: userGeminiKey // from Settings
})

// Use any AI function - automatic fallback
const summary = await hybridAI.summarize(pageText)
const embedding = await hybridAI.embed(summary)
const translated = await hybridAI.translate(summary, 'spanish')
const rewritten = await hybridAI.rewrite(summary, { 
  tone: 'professional',
  length: 'short'
})

See hybrid-ai.ts for implementation.

✅ Chrome AI: 100% offline, on-device processing
✅ Gemini API: Only summaries sent (not full page content)
✅ Local Storage: All data in IndexedDB on your device
✅ No Analytics: Zero tracking or user profiling
✅ Open Source: Auditable codebase

• AES-GCM Encryption: Summaries encrypted with 256-bit keys
• Secure Key Storage: Gemini API key in chrome.storage.sync
• Content Security Policy: Strict CSP in manifest
• HTTPS Only: Gemini API calls over secure connection
• Input Validation: Sanitization of all user inputs
• XSS Protection: Svelte's auto-escaping

• Max 10,000 memories (configurable)
• Auto garbage collection when limit reached
• Manual "Clear All" option in popup

Contributions welcome! Please follow these steps:

1. Fork the repository
2. Create a feature branch (git checkout -b feature/amazing-feature)
3. Commit changes (git commit -m 'Add amazing feature')
4. Push to branch (git push origin feature/amazing-feature)
5. Open a Pull Request

• Follow existing code style
• Add tests for new features
• Update documentation
• Keep commits atomic and descriptive

MIT License - see LICENSE file for details

• Chrome Built-in AI team for Gemini Nano APIs
• Dexie.js for excellent IndexedDB abstraction
• Svelte team for reactive simplicity
• TailwindCSS for utility-first styling

• Author: Sukhtumur N
• Email: haruschatten@gmail.com
• GitHub: @Sukhtumur
• Demo Video: [YouTube Link]

Built with ❤️ for the Chrome Built-in AI Challenge

Turn your browser into a second brain—privately and intelligently.