How to convert JSONL to a JSON array

rows.page opens .jsonl and .ndjson files, or JSONL lines you paste with Paste data. DuckDB-WASM reads them inside your browser tab, so nothing is uploaded unless you press Share.

1. Drop the file on rows.page or choose it from disk.
2. Open the Export menu and choose JSON to download one array of objects.

The export is a flat table, not a copy of your original lines. Expect these changes:

• {"user": {"city": "NY"}} comes out as {"user.city": "NY"}: nested objects become dot-path keys, up to 8 levels deep and 200 fields per object.
• Lists stay real arrays, objects inside them included.
• A whole number in a column that also holds decimals comes back as 7.0.
• Date-time strings such as 2026-01-03T10:00:00Z are read as timestamps and written back as 2026-01-03 10:00:00.

If your code expects the original records, use jq or Python instead. A JSONL file over 1.5 GB opens only its first rows, with a note saying how many were loaded, so its export is partial.

rows.page exports the current view, so you can trim the data first. Filter with country:NL, price:>20, name:* (has a value) or -name:* (is empty), or click histogram bars in the Columns panel; a filtered view exports every matching row.

For SQL, type a statement that starts with SELECT, WITH or FROM against the table data. Quote dot-path names, as in "user.city", and add ORDER BY __rid to keep the file order. SQL results export up to 1,000,000 rows.

The -s (slurp) flag reads every line and wraps the objects in one array, nesting untouched.

jq -s '.' data.jsonl > data.json

Add -c for compact output without indentation: jq -cs '.' data.jsonl > data.json. jq holds the whole array in memory, so watch your RAM on multi-gigabyte files. If a line isn't valid JSON, jq stops with a parse error and data.json stays empty.

To check the result, compare jq length data.json with grep -c . data.jsonl, which counts the non-empty lines. The two numbers should match.

Python's built-in json module needs no install. This skips blank lines and keeps every nested object as it was:

import json

with open("data.jsonl", "r", encoding="utf-8") as infile:
    records = [json.loads(line) for line in infile if line.strip()]

with open("data.json", "w", encoding="utf-8") as outfile:
    json.dump(records, outfile)

By default json.dump escapes characters like é as \u00e9; pass ensure_ascii=False to keep them as they are, and indent=2 to pretty-print. Like jq -s, this keeps the whole list in memory.

For a file bigger than your RAM, write the array one record at a time. Each line is still parsed, so a broken line stops the script with a JSONDecodeError:

import json

with open("data.jsonl", encoding="utf-8") as infile, open(
    "data.json", "w", encoding="utf-8"
) as outfile:
    outfile.write("[\n")
    records = (line for line in infile if line.strip())
    for index, line in enumerate(records):
        if index:
            outfile.write(",\n")
        json.dump(json.loads(line), outfile)
    outfile.write("\n]\n")

DuckDB keeps nested objects as objects and writes the whole array in one statement:

COPY (SELECT * FROM read_json('data.jsonl')) TO 'data.json' (FORMAT json, ARRAY true);

It infers one schema for all lines, so a key missing from a record comes back as null. The number and date changes listed above apply here too, because rows.page runs the same engine in your browser.

Before you convert, check whether the tool that receives the file reads JSONL directly. OpenAI's Batch API takes JSONL input files, and BigQuery loads JSON data only as newline-delimited JSON. For large datasets, keep JSONL or export zstd Parquet from rows.page instead.