A browser extension that converts webpages into a semantic knowledge graph. Built as a Turborepo monorepo.

• apps/extension - Browser extension (Chrome/Firefox) for capturing web pages
• apps/backend - Python FastAPI service for processing pages, extracting entities/concepts, and building the graph
• apps/dashboard - Next.js dashboard for visualizing the knowledge graph
• packages/shared - Shared TypeScript types

• Bun (for JS/TS packages)
• Python 3.11+ (for backend)
• Docker (for PostgreSQL database)
• OpenAI API key (for LLM and embeddings)

1. Install dependencies:

# Install JS/TS dependencies
bun install

# Install Python dependencies
cd apps/backend
uv sync
cd ../..

2. Set up environment variables (REQUIRED):

cd apps/backend
cp .env.example .env
# Edit .env and add your OPENAI_API_KEY
# Get your API key from: https://platform.openai.com/api-keys

⚠️ IMPORTANT: The backend will not start without an OpenAI API key. This is required for:

• Entity and concept extraction from pages
• Vector embeddings for semantic search

3. Start everything:

bun run dev

This will:

• Start PostgreSQL database in Docker
• Run database migrations
• Start backend server (http://localhost:8000)
• Start dashboard (http://localhost:3000)
• Start extension in watch mode

If you prefer to start things manually:

# Start database
bun run db:up

# Run migrations
bun run db:migrate

# Start backend
cd apps/backend
uv run uvicorn app.main:app --reload --port 8000

# Start dashboard (in another terminal)
cd apps/dashboard
bun run dev

# Build extension (in another terminal)
cd apps/extension
bun run dev

1. Build the extension:

cd apps/extension
bun run build

2. Load in browser:
   • Chrome: chrome://extensions/ → Enable Developer mode → Load unpacked → Select apps/extension/dist
   • Firefox: about:debugging → This Firefox → Load Temporary Add-on → Select apps/extension/dist/manifest.json

1. Navigate to any webpage
2. Click the extension icon
3. Click "Capture" - the page will be:
   • Extracted (clean text via Readability)
   • Processed (entities/concepts extracted via LLM)
   • Embedded (vector embeddings generated)
   • Stored (in PostgreSQL graph database)
4. View in dashboard at http://localhost:3000
5. Search and explore the knowledge graph

.
├── apps/
│   ├── extension/        # Browser extension
│   ├── backend/          # FastAPI backend
│   └── dashboard/        # Next.js dashboard
├── packages/
│   └── shared/          # Shared TypeScript types
├── docker-compose.yml    # PostgreSQL + pgvector
├── turbo.json           # Turborepo configuration
└── package.json         # Root package.json

• Page Capture: Browser extension captures HTML, URL, and timestamp
• Content Extraction: Backend extracts clean article text using Readability
• Entity/Concept Extraction: LLM extracts named entities and abstract concepts
• Embeddings: Vector embeddings for semantic search (OpenAI)
• Graph Storage: PostgreSQL with pgvector for graph storage
• Visualization: Interactive graph visualization in dashboard
• Search: Vector search with graph traversal

• bun run dev - Start everything (database + all apps)
• bun run dev:apps - Start just the apps (assumes DB is running)
• bun run build - Build all apps
• bun run db:up - Start database
• bun run db:down - Stop database
• bun run db:migrate - Run database migrations

Create apps/backend/.env with:

OPENAI_API_KEY=your_key_here
DATABASE_URL=postgresql://postgres:postgres@localhost:5432/memory_web

MIT