CSV to JSON

CSV → JSON without hidden data loss

A CSV file has rows and cells but no native data types or nested structure. A strong converter therefore has to make explicit choices about delimiters, headers, missing cells, numeric identifiers and path reconstruction. This tool defaults to conservative behavior: it preserves leading-zero identifiers, keeps unsafe integers as strings, leaves dates untouched, and reports irregular row widths.

Quick start

• Paste CSV or drop one or more files.
• Leave Delimiter on Auto unless you already know the separator.
• Use Smart & safe typing to convert ordinary booleans/numbers while protecting long IDs and leading zeros.
• Choose a nesting mode only when your headers intentionally encode paths such as user.name, items[0].sku or /user/name.
• Review the audit and profile, then copy or download the verified JSON.

What makes the conversion safer

• Quoted CSV is parsed correctly: separators, doubled quotes and embedded line breaks inside quoted fields are preserved.
• Long integer IDs are protected: values beyond JavaScript’s safe integer range stay strings by default instead of being rounded.
• Leading zeros survive: postal codes, account IDs and product codes such as 00123 remain strings.
• Duplicate headers are visible: they are auto-renamed with deterministic suffixes and counted in the audit.
• Nested-path collisions are rejected: a header set such as a and a.b cannot silently overwrite one another.
• Output is verified: generated JSON is parsed again with number-token preservation and compared with the in-memory conversion result.

Nested header examples

id,user.name,items[0].sku,items[1].sku
00123,Ada,ABC-1,ABC-2

→
[
  {
    "id": "00123",
    "user": { "name": "Ada" },
    "items": [
      { "sku": "ABC-1" },
      { "sku": "ABC-2" }
    ]
  }
]

Output modes

JSON array is the safest general-purpose export. JSON Lines writes one object per line for pipelines and large datasets. Wrapped object puts the array under a root key. Keyed object uses one unique source field as the object key and rejects duplicates.

Limits and non-goals

CSV has no universal schema, so type inference is necessarily heuristic. Currency strings, localized decimal commas, dates and mixed-format IDs are intentionally not aggressively coerced. The tool does not execute spreadsheet formulas, macros or code. Very large files still depend on browser memory; for multi-gigabyte ETL jobs, a streaming desktop or server pipeline is more appropriate.

Privacy

Conversion runs in the browser. The tool does not need to upload CSV contents to a conversion service. Batch ZIP files are assembled locally as well.