Dev Knowledge

Build knowledge base that scrapes developer documentation and makes it searchable using AI embeddings.

Turns documentation websites into a searchable database:

• Scrapes Node.js, TypeScript, Python, and JavaScript docs
• Converts HTML to clean Markdown
• Creates vector embeddings for semantic understanding
• Stores everything in a local SQLite database
• Searches using natural language queries

npm install
npm run build

npx tsx src/index.ts
# or
node dist/index.js      # after building

This will:

1. Initialize the embedding encoder
2. Scrape configured documentation sources
3. Convert HTML to Markdown
4. Generate vector embeddings
5. Store in SQLite database

Edit src/Processor.ts to modify scraping sources:

export const scraperSchema: ScrapeSchema[] = [
  {
    url: ['https://nodejs.org/docs/latest-v24.x/api/'],
    parse: (content: string): string | null => {
      // Custom parsing logic
    }
  }
  // Add more sources...
]

src/
├── core/                    # Core functionality
│   ├── Database.ts          # SQLite + sqlite-vec vector storage
│   └── embedding/           # Vector operations
│       ├── Encoder.ts       # Text to vector conversion
│       └── Decoder.ts       # Vector similarity search
├── interfaces/              # TypeScript type definitions
├── utils/                   # Utility classes
│   ├── Scraper.ts           # Web scraping with memory management
│   ├── Logger.ts            # Logging utilities
│   └── Generator.ts         # ID generation
├── index.ts                 # Main entry point
└── Processor.ts             # HTML to Markdown + embedding pipeline

• @neabyte/fetch - HTTP client with retry logic
• @xenova/transformers - Vector embeddings
• better-sqlite3 - SQLite database
• sqlite-vec - Vector similarity search
• turndown - HTML to Markdown conversion
• jsdom - DOM parsing

CREATE TABLE embedding (
  id TEXT PRIMARY KEY,
  source TEXT NOT NULL,
  content TEXT NOT NULL,
  vector BLOB NOT NULL,
  timestamp INTEGER NOT NULL
)

This project is licensed under the MIT license. See the LICENSE file for more info.