Schema Generator

How it works

• Everything runs client-side — no network request, no upload, no server round-trip. Your data never leaves the tab.
• Each column's type is inferred by scanning every value in that column (up to the first 20,000 data rows) and finding the narrowest type that fits all of them — e.g. a column of whole numbers becomes INTEGER, but if one value has a decimal point the whole column widens to a floating-point type instead of guessing a precision that might truncate real data.
• Recognized types: INTEGER/BIGINT (by magnitude), floating-point numbers, BOOLEAN (only exact true/false values — 0/1 is left as a number, since that's genuinely ambiguous), DATE (YYYY-MM-DD), TIMESTAMP (ISO 8601), and VARCHAR(n)/TEXT for everything else.
• A column named exactly id gets marked PRIMARY KEY only if every value in it is a unique, non-empty integer — turn off "Detect primary key" if that's not what you want.
• NOT NULL is only added to columns with zero empty cells in the scanned rows. A single blank cell anywhere in a column makes the whole column nullable.
• Column names are normalized to snake_case by default and de-duplicated if two headers collide; disable that to keep your original header text (still safely quoted if it contains spaces or a reserved word).
• This is heuristic inference, not a substitute for reviewing the generated DDL — always check it against what you actually intend to store before running it against a real database.