JSON to JSONL Converter

updated 28 April 2026

100% client-side. Your data never leaves the browser.

Before you start

You need a top-level JSON array — something that opens with [ and closes with ]. Each item inside the array becomes one line in the output. That item can be an object, a number, a string, even a nested array; whatever it is, I will JSON.stringify it onto a single line.

If your file is a single bare object ({...}) instead of an array, this tool can't help directly — JSONL is fundamentally a stream of records, so you'll need to wrap it: [{...}]. If your file is already JSONL, you're done; you don't need this tool.

There is no hard size limit beyond your browser's memory. The whole array gets parsed into a JS value first, so for very large files (over ~200 MB) I'd recommend a CLI like jq -c '.[]' input.json > out.jsonl instead.

Why JSONL is everywhere now

JSONL (sometimes called NDJSON) has quietly become the default for one specific reason: LLM fine-tuning. Both OpenAI and Anthropic require fine-tune training files in JSONL format — one chat example per line. If your training data starts life as a JSON array exported from a database or a labelling tool, you need to "unwrap" it before uploading. That's literally what this page does.

Beyond fine-tunes, JSONL shows up in:

• BigQuery / Snowflake / Athena bulk loads, where each line gets parsed independently so the loader doesn't have to hold the entire file in memory.
• Logging pipelines — Fluent Bit, Vector, and Logstash all emit and consume NDJSON natively.
• Kaggle and HuggingFace datasets, where streaming row-by-row beats loading a 4 GB JSON array into pandas.
• Embeddings exports from OpenAI, Cohere, and Voyage — typically one vector per line.

How to use it

1. Paste your JSON array into the Input pane, or drop a .json file onto the page.
2. Click JSON array → JSONL. I parse the whole array, then stringify each element onto its own line.
3. Check the status bar for the line count, or the error pane if something failed to parse.
4. Click Copy or Download .jsonl to save the result.

Example

Input (JSON array):

[
  {"id": 1, "user": "Ada"},
  {"id": 2, "user": "Linus"}
]

Output (JSONL):

{"id":1,"user":"Ada"}
{"id":2,"user":"Linus"}

Notice that the output is minified on each line — that's what every JSONL consumer expects. Trailing newline included so cat file1 file2 > combined still works.

Fine-tune dataset workflow

If you exported a chat dataset from a labelling tool as a JSON array, the typical pipeline looks like:

1. Paste the array here, click convert. Now it's JSONL.
2. Run the result through the OpenAI Fine Tune Validator to catch missing messages arrays, bad roles, or empty assistant turns.
3. Upload to OpenAI's files endpoint and start your fine-tune job.

I built these three tools to be used in sequence — they live in the same browser tab and don't need anything installed.

Tips & common pitfalls

• Trailing commas will break the parse. Standard JSON.parse is strict. If your array ends with ..., ], fix it first.
• The output is minified per line. JSONL consumers don't tolerate pretty-printed records that span multiple lines — every line break in the output marks a new record. If you need pretty objects, use the Formatter after this step.
• Each line is independent JSON. That means a single broken record won't poison the rest of the file the way it would in a giant array — that's the whole point of JSONL for big datasets.
• Strings, numbers, and arrays are all valid lines. JSONL doesn't require objects. If your input is [1, 2, 3], the output will be three lines: 1, 2, 3.

Troubleshooting

Related tools

• JSONL to JSONThe reverse direction — wrap one-record-per-line into a single JSON array.
• OpenAI Fine Tune ValidatorValidate the JSONL you just produced against the chat-completion schema.
• JSONL ValidatorFind malformed lines after conversion, with exact line numbers.
• CSV to JSONLIf your data started life as CSV instead of JSON, start there.

See also: if you need to do something adjacent on this site, try JSONL to CSV to flatten your fresh JSONL into a spreadsheet, Formatter to pretty-print or minify each JSONL record, or Viewer to inspect the converted records before uploading.

Frequently asked questions