File input (CSV / Excel / Parquet / JSON / Text)

File input

Load data from a dataset you've uploaded into the flow. This is usually the first node in a pipeline — the entry point for a file-based ETL flow.

File Input

uploaded dataset + format

input

Drop Nulls

remove bad rows

clean

Group By

aggregate

transform

File Output

efficient export

output

Ciaren now uses one uploaded-file input node with a format selector:

Node type	Format setting	Reads
`fileInput`	`csv` / `tsv`	Delimited values
`fileInput`	`excel`	`.xlsx` / `.xls` workbooks
`fileInput`	`parquet`	Columnar binary format
`fileInput`	`json` / `jsonl`	JSON array/records or JSON Lines
`fileInput`	`text`	Plain text, one row per line → single `text` column

Legacy csvInput, excelInput, parquetInput, jsonInput, and textInput nodes still run in existing flows, but new flows should use File Input.

Use cases

Start a pipeline from a CSV, Excel, Parquet, JSON, or plain-text file you uploaded.
Pin a flow to a specific dataset version so scheduled runs stay reproducible even after the file is re-uploaded.

Configuration

Config key	Type	Required	Description
`dataset_id`	string	Yes	The dataset to read
`dataset_version`	int	No	Pin a specific version (defaults to latest)
`format`	string	Yes	How to read the file: `csv`, `tsv`, `excel`, `parquet`, `json`, `jsonl`, or `text`

The config form only offers datasets compatible with the selected format. Changing the file type clears the selected dataset so you cannot accidentally run a CSV dataset as Parquet, for example.

Generated Python code

# fileInput with format="csv"
df_1 = pd.read_csv("sales.csv")

# fileInput with format="excel"
df_1 = pd.read_excel("report.xlsx")

# fileInput with format="parquet"
df_1 = pd.read_parquet("data.parquet")

# fileInput with format="json"
df_1 = pd.read_json("records.json")

# fileInput with format="text" — one row per line, column named "text"
df_1 = pd.read_csv("log.txt", sep="\n", header=None, names=["text"], engine="python", dtype=str)

Tips & common mistakes

Pin a version for repeatable runs. Leaving dataset_version empty always reads the latest upload; pinning makes a flow deterministic. See Dataset versioning.
No datasets listed? Upload one first on the Datasets page — the form warns when no compatible dataset exists for the node's format.
JSON shape matters. pd.read_json expects a JSON array ([{...}, {...}]) or a records-oriented object. Nested structures may require a downstream Calculated Column or custom step to flatten.
Text input produces one column. The resulting dataframe has a single text column; use String Transform or Split Column to extract structure from each line.