Convert CSV to JSON or JSON Lines

The Export menu offers both. JSON writes one array, [{...}, {...}], which is what most APIs, fixtures and front-end code expect. JSONL writes one object per line with no surrounding array, which streams, appends and splits cleanly, and is the format for fine-tuning files and log pipelines.

Same keys, same values, same order of rows. The only difference is the framing, so pick whichever the receiving side parses.

DuckDB reads the CSV and detects a type per column: integers, decimals, booleans, dates and timestamps. In the JSON, 42 is a number, true is a boolean, an empty cell is null, and dates are ISO strings like 2019-04-01. Files up to 50 MB are sniffed in full; bigger ones are sampled.

If detection can't decide, every column is read as text and a note at the top of the table says so. To force a type, run SELECT CAST(code AS VARCHAR) AS code, * EXCLUDE (code) FROM data before exporting; the JSON follows the query result.

Export writes the current view, so filter to status:active or amount:>100, hide the columns the consumer doesn't need, sort, and the file follows. Rename keys with SQL: SELECT id AS user_id, email FROM data.

Flat CSV, nested JSON? DuckDB builds objects: SELECT id, struct_pack(city := city, zip := zip) AS address FROM data exports as {"id": 1, "address": {"city": "Delft", "zip": "2611"}}. list_value(a, b) builds arrays the same way.

JSON repeats every key on every row. The CO2 sample is 13.7 MB as CSV, 109 MB as a JSON array and 109 MB as JSONL, roughly 8 times the CSV. Nothing is wrong; that is the format. If the file feeds a program rather than a person, Parquet from the same menu is 4.3 MB.

The browser ceiling is 1.5 GB per input file. SQL results are capped at 1,000,000 rows, plain filters are not.

Press Share to get a link that opens the data as a table, or run curl -T data.csv https://rows.page from a script. Anonymous links take files up to 100 MB and are deleted 7 days after creation; a free account keeps 10 datasets. The original file is served at /<id>/raw.csv with CORS headers, so a page can fetch it.

Pushing again to the same link makes a new version and the page shows how many rows are new, gone or changed.