SQL Formatter

How to use it

1. Pick the dialect that matches your database — a banner proposes one when backticks, TOP n or $$ markers give it away, and nothing switches until you accept.
2. Paste SQL or drop a .sql file; files are read locally, and anything over the 10 MB cap is refused by name rather than truncated.
3. Tune keyword case, indent width or tabs, blank lines between statements and the inline-list width — each control previews its effect before you commit.
4. Copy the result, download it as a .sql file, or open the Minify tab for a one-line version with its size readout.

Paste tangled SQL and read it back one clause per line. This page runs the sql-formatter library across 13 dialect grammars, uppercases keywords on request, and never rewrites what your quotes protect — string literals return byte-for-byte. A second tab minifies scripts to a single line and reports the byte saving. Both passes run inside a Web Worker on your device; your SQL is never uploaded.

How it works

Under Format, your text goes to the sql-formatter JavaScript library at a fixed, pinned version, so the same input keeps printing the same bytes visit after visit. The library tokenizes SQL rather than truly parsing it: it recognizes keywords, literals, comments and operators, then reprints that stream with fresh line breaks, indentation and casing. Tokenizing is also why this page cannot judge correctness — select * frmo users; formats without complaint, because to a tokenizer frmo is just another identifier.

Each of the 13 dialects loads its own grammar: MySQL treats backticks as identifier quotes and # as a comment marker, T-SQL accepts [bracketed names], PostgreSQL adds dollar-quoted strings such as $body$ … $body$. That is why the selector matters — a bracket-quoted column is a parse error under Standard SQL yet perfectly legal under T-SQL. Three cheap signals (backticks, TOP n, $$) drive the suggestion banner, which only ever asks.

The formatting library is heavy, so it loads as its own script chunk inside a Web Worker the first time you run it — off the page you first see and off the main thread while it works. Minify never touches that library at all: an in-house scanner splits the script into code, string and comment spans, collapses whitespace in code spans only, and copies every string span untouched. Neither pass performs a network request carrying your SQL.

Examples

select u.id, u.email, p.title, p.created_at from users u inner join posts p on p.user_id = u.id where u.active = 1 and p.created_at > '2025-01-01' order by p.created_at desc limit 20;

SELECT
  u.id,
  u.email,
  p.title,
  p.created_at
FROM
  users u
  INNER JOIN posts p ON p.user_id = u.id
WHERE
  u.active = 1
  AND p.created_at > '2025-01-01'
ORDER BY
  p.created_at DESC
LIMIT
  20;

create function audit_note() returns trigger as $body$
begin
insert into audit(note) values (new.id || ' changed');
return new;
end;
$body$ language plpgsql;

CREATE FUNCTION audit_note () returns trigger AS $body$
begin
insert into audit(note) values (new.id || ' changed');
return new;
end;
$body$ language plpgsql;

SELECT
  u.id,
  u.email,
  p.title,
  p.created_at
FROM
  users u
  INNER JOIN posts p ON p.user_id = u.id
WHERE
  u.active = 1
  AND p.created_at > '2025-01-01'
ORDER BY
  p.created_at DESC
LIMIT
  20;

SELECT u.id, u.email, p.title, p.created_at FROM users u INNER JOIN posts p ON p.user_id = u.id WHERE u.active = 1 AND p.created_at > '2025-01-01' ORDER BY p.created_at DESC LIMIT 20;

select id, 'DROP TABLE users' as attempted_command from audit_log where command like '%drop table%';

SELECT
  id,
  'DROP TABLE users' AS attempted_command
FROM
  audit_log
WHERE
  command LIKE '%drop table%';

Every example on this page is executed against the same engine that powers the tool, on every build.

Common errors

Parse error: Unexpected "'abc from " at line 1 column 8.

An opening quote never found its partner, so the tokenizer swallowed the rest of the statement as one string and gave up. The position points at the quote itself — close it, or double it as '' if you meant a literal apostrophe.

Parse error at token: ; at line 1 column 25

The grammar ran out of legal continuations at the named token — here a semicolon arriving while count( was still open. The actual slip usually sits left of the caret; the reported position is where recovery became impossible.

Input is larger than the 10 MB limit for this tool.

A typed refusal, not a crash, and never a silent truncation — the editor keeps your full paste. Split the dump, or run the printed npx command, which has no such ceiling.

input contains an unpaired surrogate; it was encoded as U+FFFD

The paste carried half of a two-unit character — typical of a file sliced mid-emoji or decoded with the wrong charset. Output stays well-formed; search it for � to find the repaired spot.

Keyboard shortcuts

⌘/Ctrl + Enter

Run the current mode — the explicit path once inputs get big.

⌘/Ctrl + S

Download the output as a .sql file.

⌘/Ctrl + C

Copy the whole result when no text is selected.

⌘/Ctrl + Shift + C

Copy the whole result even mid-selection.

t

Switch between light and dark theme.

? or ⌘/Ctrl + /

Open the shortcut reference dialog.

Esc

Close the open dialog.