Input
DuckDB Input
Runs a SQL query with DuckDB over one local file — Parquet, CSV, JSON (the view data) or a DuckDB database (its own tables) — and turns the result into rows.
When to use it
Use it for big analytical files: summarise a multi-million-row Parquet file before it ever becomes rows, or read a .duckdb database someone hands you.
In a pipeline
DuckDB InputChart
Settings
| Setting | Type | Required | Default or hint |
|---|---|---|---|
| File | text | Yes | sales.parquet — same places as File Input |
| Query | textarea | — | SELECT * FROM data |
| Row Limit | number | — | 1000000 |
Example
Totals per region from Parquet
- File=sales.parquet
- Query=SELECT region, SUM(total) AS total FROM data GROUP BY region ORDER BY total DESC
| region | total |
|---|---|
| North | 182400 |
| South | 96150 |
Worth knowing
- Parquet, CSV/TSV/TXT and JSON/JSON Lines files are the view data; a .duckdb (.db, .ddb) file is opened read-only and you query its tables by name.
- Only one read-only statement runs (SELECT, WITH, FROM…, DESCRIBE, SUMMARIZE); DuckDB is locked to the file's folder with no network and no extensions, so nothing is ever written.
- The file can be on the server, in My files, in Google Drive or on a computer the agent shares — remote files are fetched first.
- Large integers and dates come back as text; Type Conversion turns them back into numbers and dates. The Row Limit (default 1,000,000) stops a query that returns more.