How to convert a JSON array to JSONL

Drop your .json file onto rows.page, choose it from disk, or click Paste data and paste the text. DuckDB-WASM reads it inside your browser tab, so nothing is uploaded unless you press Share. Rows load in pages of 100 as you scroll.

1. Open the .json file on rows.page, or paste the JSON.
2. Optional: filter, sort, hide columns or run SQL in the query bar above the grid to choose what gets exported.
3. Open the Export menu and choose JSONL. You get one JSON object per line.

The file doesn't have to be a bare [...] array. If it is one object such as {"data": [...], "next": "page2"}, rows.page finds the biggest list up to 4 levels deep, opens its items as rows and shows the note: Showing the N items in "data"; the other top-level fields were skipped.

Check what your next tool expects before you export. rows.page flattens nested objects into dot-path columns such as user.address.city, up to 8 levels deep and 200 child fields per object, and the JSONL export keeps those as flat keys like "user.address.city":"Utrecht". Arrays stay arrays.

Use jq when every nested object must keep its braces, or when the file is over 1.5 GB, which rows.page refuses with "Too big for this browser". .[] walks the top-level array and -c prints each item on one compact line:

jq -c '.[]' data.json > data.jsonl

API responses often wrap the array in an object next to paging fields. Point jq at that key first, so only the array items become lines:

jq -c '.data[]' data.json > data.jsonl

For {"data": [{"id": 1, "user": {"city": "Utrecht"}}, {"id": 2, "user": {"city": "Ghent"}}], "next": "page2"} that writes 2 lines, nesting intact:

{"id":1,"user":{"city":"Utrecht"}}
{"id":2,"user":{"city":"Ghent"}}

Plain jq reads the whole file into memory before it prints a line. For a file that doesn't fit, add --stream: jq then reads the file as a stream of path and value events, and fromstream(1|truncate_stream(inputs)) rebuilds each array item and prints it as soon as it is complete. The output matches jq -c '.[]' line for line.

jq -cn --stream 'fromstream(1|truncate_stream(inputs))' data.json > data.jsonl

Use rows.page when you want to look at the data, drop bad records or reshape it with SQL before you write the .jsonl file. Use jq when the structure must stay exactly as it is or the file is over 1.5 GB.

The Export menu writes the current view: the visible columns, the active filter and the sort order. In the query bar, country:NL matches a value, price:>20 compares numbers, name:* keeps rows where the field has a value and -name:* keeps rows where it is null. Clicking a histogram bar or a top value in the Columns panel adds a filter too, and a filtered view exports every matching row.

To rename fields, add columns or remove duplicates, type SQL that starts with SELECT, WITH or FROM in the same bar and press Enter. The table is called data, and flattened names with dots need double quotes, such as "user.city":

SELECT DISTINCT * EXCLUDE (__rid) FROM data WHERE "user.city" = 'Amsterdam' ORDER BY __rid

The hidden __rid column holds the row number, so EXCLUDE (__rid) keeps it out of the duplicate check and ORDER BY __rid keeps file order. SQL results export up to 1,000,000 rows; filtered views have no cap.

A JSON array wraps every record in one outer [ and ] with commas in between, so a reader has to track that structure to split the records. JSONL, also called NDJSON or JSON Lines, puts one complete JSON value on each line, so tools can read, split and append records line by line.

• OpenAI's Batch API starts from a .jsonl file with one request per line.
• BigQuery loads JSON only as newline-delimited JSON, one object per line.
• Log pipelines append one event per line, so a file grows without rewriting a closing bracket.