Run SQL on a Parquet File
Published August 19, 2026
Run DuckDB SQL on a local .parquet file in the browser—when that is enough, and when a warehouse or CLI is the better tool.
You want SQL, not a screenshot of rows
A grid is enough to confirm the file opens. It is not enough when you need a filter, a group-by, or a join. That is SQL.
You do not need to stand up a database to get there.
Run SQL in the browser tab
Parquet Fox embeds DuckDB compiled to WebAssembly. The engine runs in this tab. Open a .parquet file, then type SQL against it. The file is not uploaded to a Parquet Fox server for open, query, or export.
Typical flow:
- Open Parquet Fox and drop the file.
- Write a
SELECTwith aLIMITfirst so you do not paint a million rows. - Add filters, aggregates, or joins once the preview looks right.
- Export the result if you need CSV, JSON, or another Parquet file.
Caps we enforce: Parquet up to 500 MB; CSV and JSON up to 2048 MB compressed on disk. Memory still depends on what you query. A SELECT * on a wide file can stall the tab even when the compressed size is legal.
When the tab is the wrong SQL tool
Use the DuckDB CLI, a Python script, or a warehouse when you need scheduled jobs, many remote tables, or a file that does not fit device RAM. The browser is a peek and a filtered export, not a cluster.
For the method table (pandas, DuckDB CLI, desktop, browser), see How to Read a Parquet File.
What to do next
- Open a file and run
SELECT * FROM the_table LIMIT 20. - If you are choosing among Python, CLI, and browser, use the pillar guide above.
FAQ
- Can I run SQL on a Parquet file without a database server?
- Yes. A browser-local viewer embeds DuckDB as WebAssembly and registers the file as a table in the tab. You do not need Postgres or a warehouse for a one-off query.
- Does the SQL engine upload my Parquet file?
- In Parquet Fox, no. Query execution runs in the tab. The file is not copied to a Parquet Fox server as part of running SQL.