CSV Input

Input

Read a CSV file and emit one row per data line. Configure the column delimiter, choose whether the first row is a header, and skip blank lines. CSV Input streams rows as it parses, so it handles large files without loading everything into memory.

Local processing

This transform runs entirely in your browser. No data is sent to external servers.

How it works

CSV Input is a source node — it produces rows from an uploaded file but takes no input handle. Parsing uses PapaParse line-by-line, which means rows are emitted as the file is read; downstream streaming transforms can begin work before the upload finishes.

The parser samples the first five data rows to infer column types. If every non-empty value in a column parses as a number, the column is coerced to number; otherwise values stay as strings. Columns whose first five rows are all empty fall back to string. Type detection happens once per column on the sample window — values past row five are coerced using the type chosen from the sample.

When Has Header Row is unchecked, the parser generates synthetic column names col_0, col_1, … and treats every line as data.

Input: A .csv file uploaded by the user. Output: A row stream where each row is an object keyed by header (or col_N).

Options

Option	Type	Default	Description
delimiter	string	,	Single character that separates columns. Common values: ,, ;, \t, |.
hasHeader	boolean	true	Treat the first non-empty row as column names. When false, columns are named col_0, col_1, …
skipEmptyLines	boolean	true	Skip lines that are empty after trimming. When false, blank lines parse as zero-column rows.

Examples

Standard comma-separated file with header

A typical sales export — comma delimiter, first row contains column names.

Before (raw file content):

order_id,customer,amount
1001,Acme Corp,2400
1002,Beta Inc,750
1003,Gamma LLC,1200

Configuration: delimiter: ",", hasHeader: true, skipEmptyLines: true.

After:

order_id	customer	amount
1001	Acme Corp	2400
1002	Beta Inc	750
1003	Gamma LLC	1200

order_id and amount are inferred as number; customer stays a string.

Semicolon delimiter, no header

A European-locale export with no header row. Synthetic column names are generated.

Before (raw file content):

2024-06-01;Alice;125.50
2024-06-02;Bob;88.00
2024-06-03;Carol;212.75

Configuration: delimiter: ";", hasHeader: false.

After:

col_0	col_1	col_2
2024-06-01	Alice	125.50
2024-06-02	Bob	88.00
2024-06-03	Carol	212.75

Tips and Edge Cases

• Type inference is sample-based, not full-scan. Numeric columns are detected from the first five non-empty rows. If row 7 of a column inferred as number contains "N/A", it will be coerced via Number("N/A") and become NaN rather than reverting to string. Pre-clean numeric columns or coerce types downstream if your file has sparse non-numeric rows past the first five. See apps/web/src/transforms/input-csv/logic.ts:45-70.
• delimiter must be exactly one character. The Zod schema rejects multi-character delimiters at config-validation time. To parse files with multi-character separators, replace the separator upstream or pre-process the file. See apps/web/src/transforms/input-csv/logic.ts:24-29.
• PapaParse runs per line, not over the whole file. Quoted values that span multiple lines are not preserved — each physical line is parsed independently. CSVs with embedded newlines inside quoted fields will produce extra rows. See apps/web/src/transforms/input-csv/logic.ts:101-109.

Related Transforms

• Excel Input — read .xlsx instead of CSV; the rest of the pipeline is identical.
• JSON Input — read structured JSON or NDJSON when the source is hierarchical.
• Change Type — fix up columns whose inferred type does not match the data.