Generate JSON Schema from Complete JSON Examples

Analyze documents before generating

Paste a complete JSON document into each editor or open a local UTF-8 .json file. Add up to twenty samples, choose Analyze samples, review the fields, then select Generate schema.

Every sample must be valid strict JSON. Fix empty editors, duplicate keys, comments, JSONC/JSON5 or numbers that lose precision before continuing; the tool does not silently skip a failing sample.

An array in one editor is one sample, and all of its elements contribute to the schema. Multiple editors produce one schema for their documents. A sample can also be a string, number, boolean or null.

The array example is:

[{"id":1},{"id":2,"nombre":"Ana"}]

With inferred required fields, the complete array passes this schema:

{
  "$schema": "http://json-schema.org/draft-07/schema#",
  "type": "array",
  "items": {
    "type": "object",
    "properties": {
      "id": {
        "type": "integer"
      },
      "nombre": {
        "type": "string"
      }
    },
    "required": [
      "id"
    ],
    "additionalProperties": true
  }
}

Decide which fields are required

In the example, id appears in both objects, so it is required by default. nombre appears in only one and is optional. If you mark nombre as required, the first object no longer passes: the report points to /0, with required at /items/required. You can still export the schema, but it will warn that some samples do not match.

A missing property and a property set to null mean different things. In [{"x":null},{"x":"a"},{}], x is optional and accepts either null or a string. Remove the empty object, and x becomes required while still accepting null.

Required fields are inferred from the examples you supply, not from your application’s rules. For nested objects, only objects at the same location are compared. Review the choices before using the schema. Resetting inferred required fields clears all your manual choices, including those on other pages.

An empty array [] produces type: "array" with items: {}, because it gives no evidence about its elements. An empty object {} reveals no properties. Integers and fractions combine as number; other mixed types form a type union. The schema can therefore accept combinations that did not occur in your examples.

What the generated schema checks

The output uses JSON Schema draft-07. It describes types, object properties, required fields and array items. Objects allow extra properties through additionalProperties: true. You can add a root title and description.

It does not infer enums, numeric ranges, patterns, references, tuples or conditions using anyOf, oneOf or allOf. Add those rules yourself if your application needs them.

Optional format suggestions can identify ASCII email addresses, absolute HTTP(S) URIs and dates written as YYYY-MM-DD. Enable them before analyzing. At least two string observations must match; a counterexample or a string over 1,024 UTF-16 units removes the suggestion. Suggestions do not add a format rule to the schema, and the built-in checker does not validate formats.

Exporting and working within the limits

Copy or download exports the complete schema, regardless of which field page you are viewing. The download is schema.json, with MIME type application/schema+json, in UTF-8 without BOM. You can choose two or four spaces of indentation. Line endings are LF, with no final newline. If clipboard access fails, copy from the full read-only output.

Limit	Maximum
Samples	20
File size	1,000,000 bytes
Text per sample / total	250,000 / 500,000 UTF-16 units
Nested containers per sample	20 levels
Values per sample / total	25,000 / 50,000
Inferred properties	2,000
Schema text	100,000 UTF-16 units
Title / description	256 / 2,000 UTF-16 units
Analysis or generation time	8 seconds

Complex samples or very long field names can reach processing limits sooner. A failed or canceled operation produces no partial schema. The field list shows up to 25 entries per page.

Files must use UTF-8. One initial BOM is removed with a notice; a second is invalid. Editing the samples or format suggestions requires a fresh analysis and resets your required-field choices. Changing required fields or schema options requires generating the output again.

Processing stays in a local Worker and requires JavaScript. Samples are not uploaded or stored. To use the related validator, copy and paste the schema yourself; links do not transfer your data.