How to query a JSON file with SQL, no database needed

Drop a .json, .jsonl, .ndjson or .ldjson file onto rows.page, or click Paste data to paste JSON or JSONL from your clipboard. A single top-level object such as {"data": [...]} is unwrapped to its biggest list, up to 4 levels deep, and a note says the other top-level fields were skipped.

1. Open the file. Rows load in pages of 100 as you scroll.
2. Type a query into the filter box above the grid, such as SELECT * FROM data WHERE price > 20. The table is always data, and the statement must start with SELECT, WITH or FROM.
3. Press Enter to run it; Shift+Enter adds a line break. Results show up to 1,000,000 rows, marked first 1M when the cap is hit.

The data table also has a hidden __rid column with the row number. ORDER BY __rid keeps the file order, and SELECT * results hide it, but SELECT DISTINCT * and SUMMARIZE still include it unless you write * EXCLUDE (__rid). SELECT DISTINCT does not keep the original row order.

rows.page flattens nested objects into dot-path columns such as user.address.city, up to 8 levels deep and 200 child fields per object. A text column holding JSON is expanded too when 90% or more of its values in the first 1,000 rows are JSON objects. Wrap these column names in double quotes in SQL: unquoted, DuckDB reads user.address.city as a column of a table called user and fails.

Arrays stay list columns, and the grid shows their first 6 items plus +N. Call unnest() in the SELECT list to give each item its own row:

SELECT
  id,
  "user.address.city" AS city,
  unnest(items) AS item
FROM data
WHERE "user.active" = true

To read one record without SQL, click its row. A side drawer opens with a JSON/Fields toggle and Copy as JSON; there is no tree view of the whole file.

The Export menu saves the current result:

• CSV: comma-delimited, UTF-8 without a BOM, header row. Cells starting with =, +, -, @, a tab or a carriage return get a leading apostrophe so spreadsheets don't run them as formulas.
• Parquet: zstd compression.
• JSON: one array of objects; flattened fields keep dot-path keys and lists stay arrays.
• JSONL: one object per line.
• Markdown: a table of the first 1,000 rows; nulls become empty cells, | is escaped as \| and lists are written as JSON text.

A SQL result exports at most 1,000,000 rows; a filtered view without SQL exports every matching row.

Share uploads the file and opens it at a https://rows.page/<id> link with your query still applied; Pin the current view on the Findings tab saves the query there. One push is capped at 100 MB on every plan, and a JSON or JSONL file over 16 MB is converted to Parquet in your browser first when that makes it at least 20% smaller. Pro ($12 per month) and Max ($39 per month) accept files up to 500 MB and 2 GB, but only through the API's multipart uploads. The link expires after 7 days unless you press Keep it on the dataset page.

rows.page refuses a .json file over 1.5 GB ("Too big for this browser") and loads only the first rows of a bigger JSONL file. It also cannot join two files. For either case, install the DuckDB CLI and query the file from your terminal:

SELECT
  user.address.city AS city,
  count(*) AS total
FROM read_json('file.json')
GROUP BY city
ORDER BY total DESC;

read_json detects both JSON arrays and newline-delimited JSON. Nested objects stay STRUCT columns instead of being flattened, so user.address.city works without double quotes.

If the JSON already sits in a database column, query it there. In Postgres, -> returns json or jsonb, so you can chain another key, and ->> returns text for filtering and sorting:

SELECT
  payload -> 'user' ->> 'city' AS city,
  count(*)
FROM events
WHERE payload ->> 'status' = 'active'
GROUP BY 1;

In SQLite, json_extract(col, '$.a.b') reads a value from a JSON text column by path:

SELECT
  json_extract(payload, '$.user.city') AS city,
  count(*)
FROM events
WHERE json_extract(payload, '$.status') = 'active'
GROUP BY 1;

• rows.page: a JSON or JSONL file up to 1.5 GB, or a public file URL, when you want SQL without installing anything, plus flattened nested fields, column histograms and export to CSV, JSON, JSONL, Parquet or Markdown.
• DuckDB CLI: a JSON or JSONL file over 1.5 GB, results over 1,000,000 rows, or a JOIN across files.
• Postgres or SQLite: the records already live in a database table.