@ngao/search-core

Lightweight search engine library for indexing and searching code, documentation, and text.

Features

• 🔍 Full-text search with ranking
• 🧠 Semantic search with embeddings
• 📝 Multi-language support (Python, JavaScript, TypeScript, Markdown, JSON, YAML)
• ⚡ Fast incremental indexing
• 💾 LanceDB vector database
• 📦 Zero external service dependencies

Installation

npm install @ngao/search-core

Pre-downloaded Models (Included)

Important: This library uses pre-downloaded local embedding models, NOT npm downloads.

The embedding models are included in the package at models/ directory:

@ngao/search-core/
├── dist/
├── models/              ← Pre-downloaded models included here
│   └── Xenova/
│       └── all-MiniLM-L6-v2/
│           ├── config.json
│           ├── tokenizer.json
│           ├── tokenizer_config.json
│           └── onnx/
│               └── model_quantized.onnx
└── package.json

For Development:

# If models are missing, download them:
cd ngao-search-core
mkdir -p models/Xenova
git clone https://huggingface.co/Xenova/all-MiniLM-L6-v2 models/Xenova/all-MiniLM-L6-v2

The library automatically detects and uses these pre-downloaded models. Benefits:

• ✅ No npm download overhead - models bundled with package
• ✅ Works offline after installation
• ✅ Faster initialization (~50ms vs ~5s download)
• ✅ Self-contained library with all dependencies

Quick Start

import { SearchEngine } from '@ngao/search-core'

const engine = new SearchEngine()
await engine.indexDirectory('./src')
const results = await engine.search('useContext')

results.forEach(result => {
  console.log(`${result.filePath}:${result.location.startLine} - ${result.name}`)
})

API Overview

SearchEngine

The main interface for indexing and searching.

interface ISearchEngine {
  indexDirectory(dirPath: string): Promise<IndexStats>
  indexFile(filePath: string): Promise<void>
  reindexFile(filePath: string): Promise<void>
  search(query: string): Promise<SearchResult[]>
  getIndexStats(): Promise<IndexStats>
}

Types

• Block - A searchable code/documentation block
• SearchResult - A search hit with relevance score
• Query - Parsed search query
• IndexStats - Index statistics

Documentation

• API Documentation
• Architecture
• Integration Guide
• Examples

License

Apache-2.0