JSONL is not valid JSON
A JSONL file cannot be handed to a standard JSON parser as a whole — there is no enclosing array and the lines are not comma-separated. Consumers must read it line by line.
Convert Parquet to JSONL
When Parquet data has to go into something that reads a stream of records — a log pipeline, a bulk loader, a training set — JSON Lines is the right target. Every row becomes one line, so the result can be processed without loading the file into memory.
- Runs in your browser
- No signup
- No upload
- Free
ParquetJSONL
Runs on your deviceCSV
Options
Reading Parquet
Auto-detect reads the first rows and picks the separator that splits them consistently.
Turn off when the file starts straight into data. Columns are then named column_1, column_2, and so on.
Turns "42" into a number and "true" into a boolean. Values with leading zeros stay text so IDs and zip codes survive.
Which worksheet to convert. Filled in once the workbook is read.
Dotted path to the array of records, such as data.items. Leave empty to use the whole document.
Turns {"user":{"id":7}} into a user.id column. Arrays are kept as JSON text.
Pick the encoding the file was written in. Wrong encoding shows up as garbled accents.
Comma-separated words that should become null instead of text.
Writing JSONL
Snappy is the default across Spark, DuckDB and pandas. Gzip is smaller but slower to read.
Turns a user.id column back into {"user":{"id":7}}.
Excel needs a byte order mark to open UTF-8 files with accents correctly.
Rows per row group. Larger groups compress better; smaller groups let readers skip more.
Converting…
Ad slot ·
conversion-preroll
This short ad is what keeps the converter free.
Inspecting Parquet records by eye?Query Parquet without exporting with ParquetReader — opens Parquet files directly.What happens
How this converter handles Parquet to JSONL
- Decompresses the Parquet row groups and reads them into records.
- Writes exactly one JSON object per line, with no wrapping array and no pretty-printing.
- Converts timestamps to ISO-8601 and oversized integers to strings so no precision is lost.
- Can rebuild nested objects from dotted column names if the downstream consumer expects structure.
Format differences
Parquet vs JSONL
Typed, compressed, columnar storage built for analytical scans. One JSON object per line, built for streaming and appending.
| Parquet | JSONL | |
|---|---|---|
| Type information | Full — ints, floats, booleans, timestamps, decimals | Strings, numbers, booleans, null |
| Structure | Columnar, with nested type support | One nested object per line |
| Compression | Built in (Snappy, Gzip, Zstd) | None |
| Typical file size | Typically 5–15% of the CSV | Slightly under JSON |
| Schema | Stored in the file footer | Implicit, per line |
| Human-readable | No — binary | Yes, one record at a time |
| Tool support | Spark, DuckDB, pandas, BigQuery, Athena | Log pipelines, LLM training sets, BigQuery |
Worth knowing
What to watch out for
The file grows substantially
Keys repeat on every line and compression is gone. JSONL is smaller than pretty-printed JSON but still many times the Parquet size.
Column order is not guaranteed to matter
JSON objects are unordered by specification. Tools that rely on key order in the output are relying on something the format does not promise.
When you need this
Common reasons to convert Parquet to JSONL
Bulk loading into BigQuery or Elasticsearch
Both accept newline-delimited JSON directly as an import format.
Streaming through command line tools
jq, grep and awk all work naturally on one-record-per-line files.
Building datasets for model training
JSONL is the de facto standard for fine-tuning and evaluation datasets.
FAQ
Parquet to JSONL questions
Is my Parquet file uploaded?
No. The conversion runs in your browser using WebAssembly and JavaScript. The file is read from your disk into a Web Worker, converted in memory, and handed back as a download. It is never sent to a server, so there is nothing for us to store, log or leak.
Should I pick JSONL or JSON?
JSONL for pipelines, imports and anything large, because it streams. JSON for a document you will open in an editor or send to an API expecting a single payload.
Does the output end with a newline?
Yes. A trailing newline after the final record is the convention, and most line-based readers expect it.
Can I get nested objects rather than dotted keys?
Yes — enable "Rebuild nested objects" in the output options.
Which codecs can be read?
Snappy, Gzip, Zstd, Brotli, LZ4 and uncompressed. Encrypted files are not supported.