Code Embedding with LanceDB (v1)

This example extracts code chunks from local Python, Rust, TOML, and Markdown files, stores the code and their vector embeddings in LanceDB, and provides a simple semantic search demo for code.

Key Features

No database setup needed: LanceDB is embedded - no external database required
Vector search: Semantic code search using sentence embeddings
Full-text search: FTS index on code content for keyword search
Portable: Data stored in ./lancedb_data/ directory - just copy to move it
Syntax-aware chunking: Uses RecursiveSplitter with language detection

Data Storage

All data is stored in the ./lancedb_data/ directory in your project folder. This directory is created automatically on first run.

To start fresh, simply delete the ./lancedb_data/ directory and re-run the indexing.

Run

Install dependencies:

pip install -e .

Build/update the index (writes rows into LanceDB). Pick one of the two modes:

Catch-up run — scan sources, sync changes, exit:
sh
```
cocoindex update main
```
Live run — catch up, then keep watching for file changes (the source declares live=True in main.py):
sh
```
cocoindex update -L main
```

Query:

python main.py "embedding"

Environment

Copy .env.example to .env and fill in the blanks — it is loaded automatically when you run the example:

cp .env.example .env

LanceDB is an embedded local store, so there are no required secrets — the file documents the optional LANCEDB_URI override.