CSV Delimiter Auto-Detect

Stop guessing. Feed in any CSV/TSV/PSV/text file, get back a single decision row with the delimiter, quote char, header presence, encoding, confidence, and top-2 alternatives.

Pure JavaScript, zero dependencies. Pairs with every other CSV Actor in the portfolio (csv-dedupe-normalizer, csv-schema-profiler, csv-coerce-types, csv-quality-scorecard, csv-join-enricher, csv-diff, csv-statistical-summary).

How it scores

For each candidate delimiter (, \t ; | by default):

1. Split the sample into rows.
2. Find the most common column count.
3. Score the fraction of rows that have that count (consistency).
4. Score the per-column type consistency (numeric / date / text uniformity).
5. Detect header presence by checking whether row 0's types differ from the rest.

Composite score = 0.5 * consistency + 0.4 * typeUniformity + 0.1 * headerBonus.

Use cases

• The single most common "I can't load this CSV" pain in n8n / Make / Google Sheets — the file uses ; (European Excel), \t (TSV), | (pipe), or , (US default) and the user doesn't know which.
• Auto-route files into the right downstream Actor: pass the detected delimiter into csv-coerce-types or csv-dedupe-normalizer.
• Audit a partner's data feed: "they keep sending PSV and we keep parsing it as CSV".

Input

Output

{
  "_kind": "decision",
  "url": null,
  "encoding": "utf-8",
  "delimiter": ";",
  "quoteChar": "double",
  "hasHeader": true,
  "rows": 4,
  "columnCount": 3,
  "confidence": 0.94,
  "alternatives": [
    { "delimiter": ",", "score": 0.41, "rows": 4, "columnCount": 1 },
    { "delimiter": "\\t", "score": 0.30, "rows": 4, "columnCount": 1 }
  ]
}

Run locally

npm install
npm test     # 10 cases

Pricing / cost expectations

Default 256 MB / 512 MB. Reads at most 64 KB of the input (configurable via sampleBytes). Pay-per-event scheduled to activate after public launch: 1 charge per delimiter_decided event.

Related Actors

• ../csv-header-normalizer — once you know the delimiter, normalize the headers.
• ../csv-schema-profiler — per-column type inference.
• ../csv-dedupe-normalizer — dedupe + header normalization.
• ../csv-quality-scorecard — full data-quality scorecard.

FAQ

Does it detect the quote char? Yes, both ' and " are scored and the best is chosen.

Does it handle UTF-8 BOM? Yes — the encoding field is set to utf-8-bom and the BOM is stripped before scoring.

Does it work on one-line samples? Single-row CSVs get a low confidence score (the detector needs at least 2 rows to be confident).