JSONL

JSONL (JSON Lines) is a format in which each line of a file is an independent JSON document, almost always an object. Lines are separated by single \n characters. There is no wrapping array, no comma between records, and no top-level structure beyond the stream of lines.

{"id": 1, "name": "Ada"}
{"id": 2, "name": "Linus"}
{"id": 3, "name": "Grace"}

The format has no separate spec beyond the short description at jsonlines.org. Each line must be valid JSON on its own, and each line should be written without internal newlines (i.e., minified).

Why the format exists

Standard JSON requires the entire file to be read before any of it can be used, because everything is nested inside one top-level value. For large datasets this is inefficient. JSONL replaces the wrapping array with a stream of records, which brings three immediate wins:

• Streaming — you can process records one at a time without holding the whole dataset in memory.
• Append-only writes — adding a record is echo '{"...": "..."}' >> file.jsonl. No splicing into an existing array.
• Parallel processing — a large JSONL file can be split on newline boundaries and each chunk fed to a separate worker.

Where it shows up

• Logs: structured logging tools (Vector, Fluentd, cloud platforms) emit JSONL by default.
• Data pipelines: Kafka Connect, BigQuery exports, PostgreSQL COPY ... FORMAT json, MongoDB mongoexport.
• ML training sets: Hugging Face datasets and many LLM providers use .jsonl for instruction tuning and evaluation data.

Gotchas

• Do not pretty-print individual records — pretty-printing inserts newlines, which destroys the one-record-per-line invariant.
• Some editors or Git on Windows convert \n to \r\n. Strict parsers reject that; lenient ones accept it.
• A BOM at the start of the file can break parsing of the first record. Always save as UTF-8 without BOM.

See the guide: JSONL, NDJSON, and JSON Lines explained.