Convert JSON to CSV, nested objects included

A flat array of objects is the easy case: every key is a column, every object is a row. Nested objects are flattened into dot-path columns, so {"address": {"city": "Rotterdam"}} gives a column named address.city. Flattening goes 4 levels deep and stops at 200 columns; what is left stays as JSON text in one column.

Arrays are not exploded into rows. An array value is written to the CSV cell as JSON text, like ["nl","en"], so a row stays a row. Strings that themselves contain JSON are parsed and flattened too, when at least 90% of a 1,000-row sample parses.

An array of objects ([{...}, {...}]), one object per line (JSON Lines, NDJSON), or a single object. A wrapper like {"items": [...]} opens as one row with the array in a cell; unpack it with SELECT unnest(items, recursive := true) FROM data and export the result.

Keys that are missing on some objects become empty cells. Types are detected per column across the file (whole file up to 50 MB, sampled above that), so 42 stays a number and "42" stays text.

Export writes the current view. Filter with region:Europe or population:>1000000, hide the columns nobody needs, sort by clicking a header, and the CSV follows. A plain filter has no row cap; SQL results stop at 1,000,000 rows.

Renaming is a one-line query: SELECT name.common AS country, capital[1] AS capital, population FROM data. The export takes the column names of the result.

UTF-8, comma separated, a header row, values quoted when they contain commas, quotes or line breaks. null becomes an empty cell, booleans are written as true and false, numbers without quotes. Text that starts with =, +, - or @ gets a leading apostrophe so a spreadsheet won't execute it.

Need the other direction, or a different output? The same Export menu writes JSONL, Parquet with zstd compression and a Markdown table of the first 1,000 rows.

The ceiling is 1.5 GB per file, because everything runs in the memory of your tab. For a repeatable job use DuckDB on your machine: duckdb -c "COPY (SELECT * FROM read_json('data.json')) TO 'data.csv' (HEADER)" does the same flattening.

To pass the data on, press Share or run curl -T data.json https://rows.page. The link opens the table for anyone, the original file stays downloadable at /raw.json, and anonymous links are deleted after 7 days unless you keep them.