CSV to NDJSON Converter

How to use

1. 1Paste your CSV into the input box, or click Load sample for a realistic example that exercises quoted fields, embedded commas, and dotted headers.
2. 2Pick the delimiter (comma, semicolon, tab, or pipe) and the header mode. Use First row is the header for normal CSVs; use No header row to auto-name columns col_1, col_2 when your file starts with data.
3. 3Choose the output shape. Keep flat keys when downstream consumers expect address.city verbatim. Switch to Rebuild nested objects to turn dotted headers and indexed columns back into nested objects and arrays.
4. 4Adjust the row width policy and the cleanup toggles: pad short rows with null or fail strict, infer types, treat empty cells as null, trim whitespace from cells, and skip blank rows.
5. 5Use the Pretty preview tab to verify the first few records look right, then click Copy NDJSON to copy the canonical text or Download .ndjson to save it as a file. The download streams one record per line, ready for BigQuery, Athena, Snowflake, OpenAI fine-tune jobs, or any tool that expects NDJSON.

About this tool

CSV to NDJSON Converter turns a CSV (or TSV, semicolon, or pipe-separated file) into newline-delimited JSON, the streaming format that BigQuery, Athena, and Snowflake ingest natively and that OpenAI fine-tuning, Anthropic batch jobs, Hugging Face Datasets, and log pipelines like Fluentd, Vector, and Loki read with one record per line. The parser implements the same RFC 4180 subset the rest of the catalog uses: quoted fields with doubled-quote escapes, multi-line cells inside quotes, mixed LF and CRLF row separators, a leading UTF-8 BOM stripped automatically, and clear error messages with line and column numbers for unclosed quotes or stray quote characters. Headers can come from the first row or be auto-numbered as col_1, col_2; empty and duplicate header cells are sanitized (an empty cell becomes col_N, a duplicate gets a numeric suffix) so the JSON is always valid. Smart type inference is conservative: only the exact tokens true, false, and null become their JSON-native types, only purely-digit strings that fit a safe JavaScript integer become numbers, only fixed-point decimals become floats, and everything else (including 19-digit Snowflake IDs and zip codes with leading zeros) stays as a string. Nested rebuild reads dotted headers and rebuilds nested objects and arrays in the output, the inverse of the flatten convention used by pandas.json_normalize and the existing NDJSON-to-CSV tool: address.city becomes a nested object, tags.0 and tags.1 become an array, items.0.name becomes an array of objects, and a backslash escapes a literal dot in a header. A row-width policy decides what to do with ragged rows: pad with null and keep going, or stop on the first mismatch and surface the row number. Empty cells can become an empty string or JSON null. The output panel offers two views (raw NDJSON for export and a pretty-printed preview of the first records for human reading), a header bar with row, column, and byte counts, a Copy button that always copies canonical one-record-per-line text, and a Download .ndjson button that streams the same text as a file with the application/x-ndjson MIME type. Everything happens locally in your browser. Your CSV never leaves your device, which matters because production data often includes personally identifiable values that you do not want to upload to a third-party tool.

Free to use. Works in your browser. No signup, no login.