How does it infer field types?

It reads type hints in your text — words like string, number, integer, boolean, date, or email next to a field name — and falls back to string when no type is stated. You can edit the generated schema afterward.

Does this call an LLM?

No. It is pure local parsing — no API key, no network request. That makes it instant and private, at the cost of being heuristic rather than fully understanding free-form prose.

What JSON Schema version does it emit?

It emits a draft 2020-12 schema with type, properties, items for arrays, and a required array. This is the dialect OpenAI and most validators expect for structured outputs.

How should I write the description for best results?

List fields clearly with a name and a type, separated by commas or new lines, for example 'name: string, price: number, in stock: boolean'. The clearer the field list, the better the inferred schema.

Output Schema Guesser

Turn a plain-English description into a JSON Schema

When you want an LLM to return structured output, you need a JSON Schema — but hand-writing one for every new shape is tedious. This tool reads a natural language description of the structure you want and drafts a JSON Schema (draft 2020-12) with the right fields, types, and a required list. No API key, no network call — it parses your text locally and instantly.

How the guesser works

The parser scans your description for field names and nearby type hints — words like string, number, integer, boolean, date, or email. Each detected field becomes a property with the inferred type (defaulting to string when no hint is present). You choose whether the root is a single object or an array of objects; for arrays the fields go under items. Field names are normalised to clean keys, and every property is listed as required by default so your structured-output call is strict. The result is a ready-to-edit schema, not a final answer — tweak types and required-ness to taste.

Tips and examples

Write one field per line with an explicit type: title: string, price: number, published: boolean. Clear lists parse far better than prose.
Mention “array of” or “list of” at the start to get an array root automatically.
Use integer when you mean whole numbers (counts, ids) and number for decimals (prices, scores) — the schema preserves the distinction.
The generated required array assumes every field is mandatory; delete entries for optional fields after copying.