Convert CSV to Nested JSON Grouped by Column

Grouped mode, group column = category. Each category becomes a key holding its product rows; price stays a number.

Input (products.csv):
name,category,price
Laptop,Electronics,999
Mouse,Electronics,25
Novel,Books,15

Output (Grouped by column: category):
{
  "Electronics": [
    { "name": "Laptop", "category": "Electronics", "price": 999 },
    { "name": "Mouse", "category": "Electronics", "price": 25 }
  ],
  "Books": [
    { "name": "Novel", "category": "Books", "price": 15 }
  ]
}

Group column = owner. Useful for a per-person view in a config or dashboard feed.

Input:
task,owner,done
Deploy,ada,false
Review,ada,true
Design,linus,false

Output (Grouped by column: owner):
{
  "ada": [
    { "task": "Deploy", "owner": "ada", "done": false },
    { "task": "Review", "owner": "ada", "done": true }
  ],
  "linus": [ { "task": "Design", "owner": "linus", "done": false } ]
}

A row with no region lands in the _unknown bucket so it is not lost.

Input:
customer,region
Acme,EMEA
Globex,

Output (Grouped by column: region):
{
  "EMEA": [ { "customer": "Acme", "region": "EMEA" } ],
  "_unknown": [ { "customer": "Globex", "region": "" } ]
}

Grouping does not fold case, so Books and books are two keys. Normalize the column first if you want them merged.

Input:
title,shelf
A,Books
B,books

Output (Grouped by column: shelf):
{
  "Books": [ { "title": "A", "shelf": "Books" } ],
  "books": [ { "title": "B", "shelf": "books" } ]
}
→ normalize case with csv-case-converter first to merge

Grouped is one level. For a parent/child tree from dotted headers, convert to an array here, then unflatten.

Input:
id,address.city,address.zip
1,Paris,75001

Step 1 (Array mode here):
[ { "id": 1, "address.city": "Paris", "address.zip": 75001 } ]

Step 2 (json-unflattener):
[ { "id": 1, "address": { "city": "Paris", "zip": 75001 } } ]

Rows whose grouping cell is empty or absent collect under a single _unknown top-level key rather than being dropped. If you see an _unknown bucket you did not expect, some rows are missing the group value — clean or fill that column in the source CSV before converting.

If you pick Grouped by column but do not select a grouping column, the output falls back to a plain array (no grouping). Always confirm a column is chosen in the Group records by column dropdown so you get the nested object you expect.

JSON object keys are always strings. If you group by a numeric column (e.g. year), inference may type the value 2026 as a number inside the record, but the top-level key is the stringified "2026". This is normal JSON behaviour, not a defect.

Books and books produce two separate keys — the grouper does not fold case or trim. If you want them merged, normalize the grouping column first with csv-case-converter (or trim with the cleaner) before converting.

The grouping column is not removed from the inner records — each row keeps its category/owner/region field even though it is now redundant with the key. If you want it dropped from the records, remove that column first with csv-column-remover, then convert.

Grouped mode buckets by exactly one column. You cannot group by category then sub-group by subcategory in a single pass. For a deeper tree, convert to an array and reshape with a script, or expand dotted headers into nested objects using json-unflattener.

Grouping is about bucketing rows under a key — it does not interpret address.city style headers as nesting. Those stay literal keys inside the records. Use the array-then-json-unflattener recipe in the cookbook to build parent/child objects from dotted headers.

Free tier caps at 2 MB / 500 rows. Grouping a large dataset (e.g. all orders by customer) may exceed it — Pro raises the cap to 100 MB / 100,000 rows. The number of distinct groups does not have its own limit; the cap is on the input size and row count.