SQL to ER Diagram

How to use it

1. Pick the dialect that wrote your DDL — quoting is what differs: PostgreSQL double-quotes identifiers, MySQL backticks them, T-SQL brackets them, SQLite accepts all three.
2. Paste CREATE TABLE statements (pg_dump --schema-only output works as-is) or drop a .sql file; files are read locally and a non-UTF-8 byte gets a visible replacement notice.
3. Read the strip above the diagram — the table and relationship counts are computed from what actually parsed, and anything skipped is itemized below the canvas.
4. Pan by dragging, zoom with the buttons or + and − on the focused canvas, then export the exact drawing as an .svg file or a .png raster at up to 2× — a diagram too large for the browser's canvas limit scales down, and the export note names the factor used.

Paste the CREATE TABLE half of a schema dump and watch it become a crow's-foot diagram: one card per table, key badges on the columns, and an edge for every foreign key with its cardinality read from your constraints. Layout, parsing and SVG/PNG export all run on this device — the DDL never leaves the tab. Statements outside the declared subset are counted and listed, never silently ignored.

How it works

First a span scan walks the paste under the selected dialect's quoting rules and marks what is string, comment or code — the same shared scanner layer the other SQL pages here build on. That pass is why a semicolon inside 'a;b', a # comment in MySQL or a PostgreSQL $body$ … $body$ block never splits a statement in the wrong place: statement boundaries only count in code spans.

Next the DDL parser reads a declared subset — CREATE TABLE with inline and table-level PRIMARY KEY, FOREIGN KEY, UNIQUE and REFERENCES clauses, plus ALTER TABLE … ADD CONSTRAINT — into a table model: columns with types, nullability and defaults, key sets, and every foreign key with its ON DELETE / ON UPDATE actions. Whatever falls outside that subset (views, indexes, functions, CHECK bodies, DML) becomes a counted skip entry with its character span, so the diagram's silence is never ambiguous.

Cardinality is then inferred from keys alone: a plain foreign key draws one-to-many; a foreign key whose columns are the child's own primary key or a UNIQUE set draws one-to-one; and a table with exactly two foreign keys and no columns of its own prints an extra dashed many-to-many edge between the two targets — a heuristic, labeled as such, because SQL has no join-table keyword.

Finally a layered layout assigns each table a column: referenced parents sit a layer before their children, cycles are broken deterministically, and every coordinate is integer arithmetic — the same DDL always yields byte-identical geometry, which the test suite asserts by running the layout twice and comparing bytes. The result is pure draw data (rectangles, paths, text positions) that the page renders as SVG elements; nothing is ever assembled from markup strings.

Examples

CREATE TABLE customers (id int PRIMARY KEY);
CREATE TABLE orders (id int PRIMARY KEY, customer_id int REFERENCES customers (id));

2 tables · 1 relationship
orders -> customers [one-to-many] customer_id

CREATE TABLE students (id int PRIMARY KEY);
CREATE TABLE courses (id int PRIMARY KEY);
CREATE TABLE enrollments (student_id int, course_id int, PRIMARY KEY (student_id, course_id), FOREIGN KEY (student_id) REFERENCES students (id), FOREIGN KEY (course_id) REFERENCES courses (id));

3 tables · 2 relationships
enrollments -> students [one-to-many] student_id
enrollments -> courses [one-to-many] course_id
students -> courses [many-to-many, heuristic] enrollments

CREATE TABLE employees (id integer PRIMARY KEY, name text NOT NULL, manager_id integer REFERENCES employees (id));

1 table · 1 relationship
employees -> employees [one-to-many, self] manager_id

CREATE TABLE users (id int PRIMARY KEY);
CREATE TABLE profiles (user_id int PRIMARY KEY REFERENCES users (id), bio text);
CREATE TABLE avatars (id int PRIMARY KEY, media_id int REFERENCES media (id));

3 tables · 2 relationships
profiles -> users [one-to-one] user_id
avatars -> media [one-to-many, dangling] media_id

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

Common errors

No CREATE TABLE statements found in this input. This looks like DML (SELECT/INSERT/UPDATE/DELETE) — paste schema DDL instead, e.g. the output of pg_dump --schema-only.

Queries describe questions, not structure — there is no table shape inside a SELECT. Export the schema itself: pg_dump --schema-only, SHOW CREATE TABLE in MySQL, or .schema in the sqlite3 shell all produce the CREATE TABLE text this page reads.

Input is 600,000 characters — over the 524,288-character (512 KiB) limit for this tool.

The refusal names both the cap and your actual size, and your paste stays in the editor untouched. Half a megabyte of pure CREATE TABLE text is an enormous schema — trim the dump to the tables you care about, or split it by schema.

This DDL defines 121 tables — more than the 120 this diagram can render.

Past 120 tables a single picture stops being readable long before it stops being computable, so the tool refuses rather than render a smear. Diagram one subsystem at a time — foreign keys to tables you left out will still show as dashed ghosts.

orders.customer_id references "customers", which is not defined in this DDL — the relationship is drawn dashed to a placeholder.

Not a failure — a disclosure. The constraint is drawn, dashed, to a ghost card, and this note names exactly which definition is missing so you know what else to paste. It disappears the moment customers is defined in the same input.

Keyboard shortcuts

⌘/Ctrl + Enter

Re-render the diagram and re-fit it to the canvas.

⌘/Ctrl + S

Download the diagram as an .svg file.

+ − 0

Zoom in, zoom out, or fit to view when the diagram has focus.

← ↑ ↓ →

Pan the focused diagram.

t

Flip this page between light and dark themes.

? or ⌘/Ctrl + /

Open the shortcut reference in an overlay.

Esc

Close the open dialog.