Skip to main content

DuckDB

DuckDB is an in-process SQL OLAP database management system.

Load a DuckDB query with one document per row.

%pip install --upgrade --quiet  duckdb
from langchain_community.document_loaders import DuckDBLoader
%%file example.csv
Team,Payroll
Nationals,81.34
Reds,82.20
Writing example.csv
loader = DuckDBLoader("SELECT * FROM read_csv_auto('example.csv')")

data = loader.load()
print(data)
[Document(page_content='Team: Nationals\nPayroll: 81.34', metadata={}), Document(page_content='Team: Reds\nPayroll: 82.2', metadata={})]

Specifying Which Columns are Content vs Metadata​

loader = DuckDBLoader(
    "SELECT * FROM read_csv_auto('example.csv')",
    page_content_columns=["Team"],
    metadata_columns=["Payroll"],
)

data = loader.load()
print(data)
[Document(page_content='Team: Nationals', metadata={'Payroll': 81.34}), Document(page_content='Team: Reds', metadata={'Payroll': 82.2})]

Adding Source to Metadata​

loader = DuckDBLoader(
    "SELECT Team, Payroll, Team As source FROM read_csv_auto('example.csv')",
    metadata_columns=["source"],
)

data = loader.load()
print(data)
[Document(page_content='Team: Nationals\nPayroll: 81.34\nsource: Nationals', metadata={'source': 'Nationals'}), Document(page_content='Team: Reds\nPayroll: 82.2\nsource: Reds', metadata={'source': 'Reds'})]

Help us out by providing feedback on this documentation page: