Schema Inferrer

Infer a JSON Schema and a SQL CREATE TABLE from sample data

Paste a sample dataset — CSV/delimited text with a header row, or JSON (one record object or an array of record objects) — and this tool infers a type and nullability for every column, then emits two schema artifacts at once:

• a standard JSON Schema (Draft 2020-12 or Draft-07), and
• a SQL CREATE TABLE statement for your chosen dialect.

It reads schemas only — no INSERT rows — so the output is exactly the structure definition you paste into a validator, a migration, or an ORM. Everything runs locally in your browser.

Worked example

Input (JSON records):

[{"id":1,"email":"[email protected]","active":true},
 {"id":2,"email":"[email protected]","active":false}]

With table users and primary key id, the both output is:

{
  "$schema": "https://json-schema.org/draft/2020-12/schema",
  "title": "users",
  "type": "array",
  "items": {
    "type": "object",
    "properties": {
      "id": {
        "type": "integer"
      },
      "email": {
        "type": "string",
        "format": "email"
      },
      "active": {
        "type": "boolean"
      }
    },
    "required": [
      "id",
      "email",
      "active"
    ],
    "additionalProperties": false
  }
}

CREATE TABLE "users" (
  "id" INTEGER NOT NULL,
  "email" TEXT NOT NULL,
  "active" BOOLEAN NOT NULL,
  PRIMARY KEY ("id")
);

How inference works

• Types: each column is checked most-specific first — boolean → integer → float → date → datetime → text. Zero-padded codes (007, ZIP codes) stay text so they keep their leading zero.
• Nullability: a column that has a value in every record becomes NOT NULL (SQL) and appears under required (JSON Schema). A blank CSV cell, a JSON null, or a key missing from some record makes the column nullable — its JSON Schema type becomes ["<type>", "null"]. Turn this off with NOT NULL for full columns.
• String formats: text columns whose values all match email, uri, uuid, or ipv4 get that JSON Schema format. Date and datetime columns always carry date / date-time. (Draft-07 has no uuid format, so it is omitted there.)
• Primary key: the named column is marked PRIMARY KEY in the SQL and treated as required / NOT NULL in both outputs.

Limits and edge cases

• One flat table only — nested JSON objects/arrays are stringified into a single text/string column rather than split into related tables (no FOREIGN KEY / normalization).
• No INSERT statements — this tool emits schemas/DDL only. For data-loading SQL, use a CSV-to-SQL converter instead.
• VARCHAR lengths are not sized to the data: text maps to TEXT (MySQL/PostgreSQL/SQLite), NVARCHAR(255) (SQL Server), or VARCHAR(255) (ANSI).
• With no data rows, types cannot be inferred and every column falls back to a nullable string / TEXT.