> ## Documentation Index
> Fetch the complete documentation index at: https://www.landbase.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# How to export data in different formats

> Choose the right output format for your use case — JSONL, CSV, Parquet, or compressed JSONL — and understand what happens during each export.

# How to export data in different formats

This guide shows you which format to use and what to expect during each export.

***

## Formats at a glance

| Format           | Extension   | Best for                                                |
| ---------------- | ----------- | ------------------------------------------------------- |
| JSONL            | `.jsonl`    | Scripts, Python, `jq` pipelines, maximum field coverage |
| Compressed JSONL | `.jsonl.gz` | Storage, large transfers, feeding into data pipelines   |
| CSV              | `.csv`      | Excel, Google Sheets, BI tools, human review            |
| Parquet          | `.parquet`  | Analytics, DuckDB, Pandas, columnar processing          |

***

## How to specify the format

Pass the desired extension to `--download`:

```bash theme={null}
# JSONL
landbase-cli "find fintech companies in NYC" --download=results.jsonl

# CSV
landbase-cli "find fintech companies in NYC" --download=results.csv

# Compressed JSONL
landbase-cli "find fintech companies in NYC" --download=results.jsonl.gz

# Parquet
landbase-cli "find fintech companies in NYC" --download=results.parquet
```

For dataset downloads, use:

```bash theme={null}
landbase-cli datasets download <dataset-id> ./output.csv
landbase-cli datasets download <dataset-id> ./output.parquet
```

***

## What happens during each export

### JSONL and CSV

Both JSONL and CSV trigger a **publish workflow** before downloading. Landbase runs the publish step asynchronously — it formats and flattens the data, then makes it available for download. This usually takes 30–90 seconds.

You will see a progress indicator while the publish step runs. The CLI polls until the file is ready.

### Compressed JSONL

Same as JSONL — triggers a publish workflow. The output is compressed with gzip. Unzip with:

```bash theme={null}
gunzip results.jsonl.gz
```

Or read without unzipping:

```bash theme={null}
zcat results.jsonl.gz | jq .
```

### Parquet

Parquet downloads the **native dataset bytes** directly — no publish step. This is the fastest format for large datasets. Use it when feeding results into DuckDB, Pandas, or Spark:

```python theme={null}
import pandas as pd
df = pd.read_parquet("results.parquet")
```

***

## Choosing between JSONL and CSV

Use **JSONL** when:

* You are writing a script or piping through `jq`
* You want every available field (CSV may truncate or flatten nested fields)
* You are loading data into Python or a database

Use **CSV** when:

* You are opening in Excel or Google Sheets
* You are handing off to someone non-technical
* A BI tool requires it

***

## Publishing a dataset you already have

If you already have a dataset from a prior run and want to download it in a new format:

```bash theme={null}
landbase-cli workflow publish <dataset-id> --format=csv --wait
CHILD=$(landbase-cli datasets lineage <dataset-id> --direction=children \
  | jq -r '[.[] | select(.workflow_type == "publish")] | last | .id')
landbase-cli datasets download $CHILD ./output.csv
```

***

## Related

* [How to chain workflows](/docs/how-to/chain-workflows) — publish as part of a full pipeline
* [datasets reference](/docs/reference/datasets) — `datasets download` flags
* [workflow reference](/docs/reference/workflow) — `workflow publish` flags