Skip to main content
OSS-only In Python, LanceDB tables can also be queried with DuckDB, an in-process SQL OLAP database. This means you can write complex SQL queries to analyze your data in LanceDB. The integration is done via Apache Arrow, which provides zero-copy data sharing between LanceDB and DuckDB. DuckDB is capable of passing down column selections and basic filters to LanceDB, reducing the amount of data that needs to be scanned to perform your query. Finally, the integration allows streaming data from LanceDB tables, allowing you to aggregate tables that don’t fit into memory.
DuckDB quacks ArrowAll of this uses the same mechanism described in DuckDB’s blog post” on how it integrates with Apache Arrow.
We can demonstrate this by first installing duckdb and lancedb.
bash
pip install duckdb lancedb
We will re-use the dataset created previously: The to_lance method converts the LanceDB table to a LanceDataset, which is accessible to DuckDB through the Arrow compatibility layer. To query the resulting Lance dataset in DuckDB, all you need to do is reference the dataset by the same name in your SQL query. 
┌─────────────┬─────────┬────────┐
│   vector    │  item   │ price  │
│   float[]   │ varchar │ double │
├─────────────┼─────────┼────────┤
│ [3.1, 4.1]  │ foo     │   10.0 │
│ [5.9, 26.5] │ bar     │   20.0 │
└─────────────┴─────────┴────────┘
You can very easily run any other DuckDB SQL queries on your data. 
┌─────────────┐
│ mean(price) │
│   double    │
├─────────────┤
│        15.0 │
└─────────────┘