Generate

Any format, through JSON, into a language

{
  "name": "Launch party",
  "seats": 120,
  "price": 12.5,
  "online": false,
  "hosts": ["Ada", "Linus"]
}

# Generated by myjsoneditor.com — Python 3.11+
# Requires pydantic>=2

from pydantic import BaseModel


class Event(BaseModel):
    name: str
    seats: int
    price: float
    online: bool
    hosts: list[str]

The row above the panes reads from left to right. From names what you are pasting — JSON, YAML, XML, CSV or TSV — and Language offers the targets, five as tabs and five more under More. A second row, Structured output, holds the three AI providers. Only one target is selected across both rows at a time.

Whatever the source is, the generator only ever reads JSON. A JSON source goes straight in; anything else is read into JSON first, and the row shows that middle step. The name above the result is the file Download saves, and Copy takes the code. In the editor, Generate in the band opens a panel titled Generate code with the same targets over the document in the pane, which is the quicker route when the document is already open there.

TypeScript interfaces from a JSON API response

TypeScript is the target the page opens on, and the one most readers want. TypeScript interfaces from a JSON API response takes a users response with two unlike records through it, and explains each decision in the output: the question marks, the names of the nested interfaces, what an empty list becomes, and how a null changes a field’s type.

The other nine languages

Every language reads the same model of the data, so a field that is optional in TypeScript is optional in Rust and nullable in Java, each in its own idiom. The other nine languages runs one short list of books through all nine, side by side, with a note on each about the choices that language makes and the options that change them.

From YAML, XML or CSV

service: billing
replicas: 3
ports:
  - name: http
    port: 8080
  - name: metrics
    port: 9090
    public: false

JSON
{
  "service": "billing",
  "replicas": 3,
  "ports": [
    {
      "name": "http",
      "port": 8080
    },
    {
      "name": "metrics",
      "port": 9090,
      "public": false
    }
  ]
}

TypeScript
export interface Port {
  name: string;
  port: number;
  public?: boolean;
}

export interface Service {
  service: string;
  replicas: number;
  ports: Port[];
}

With From set to anything but JSON, the result grows two tabs: JSON, the document the source was read into, and one named after the target, in front. The types come only from the JSON tab, so it is the place to look when a type surprises you. Here public is optional because one port lacks it, and the ports are numbers because the YAML wrote them bare.

XML and CSV are less sure about numbers than YAML is. XML stores everything as text, so an element holding 8080 becomes a string unless Detect numbers and booleans is ticked under Read XML in Options. A CSV cell in quotes stays text, so a column with one quoted number comes out as string | number. Neither reading option is remembered. Open JSON in Editor takes that JSON to the editor, for when the reading needs fixing before the types are worth keeping.

Optional fields, nested types and the root name

{
  "members": [
    { "name": "Ada", "email": "[email protected]" },
    { "name": "Linus" }
  ]
}

The defaults
export interface Member {
  name: string;
  email?: string;
}

export interface Crew {
  members: Member[];
}

name: T | undefined, and Roster
export interface Member {
  name: string;
  email: string | undefined;
}

export interface Roster {
  members: Member[];
}

Each target’s settings sit under its name in Options, and they are remembered: next week, in any tab, the page opens with the ones you chose. They are also shared with the editor’s Generate panel, so a setting changed there is changed here. And the collapsed Options row shows only TypeScript options, not their values. That is why every example in these guides names each setting, even one left at its default.

Root name is different. It belongs to one document: it is kept for the tab, and a new document clears it. Left empty, it shows the name the page proposes, taken from the file name — here Crew — while nested types are named after their keys, and an array’s items take the singular, so members gives Member. A typed name replaces the root’s, and leaves the types named after keys alone.

Types for one part of a document

{
  "data": {
    "users": [
      { "id": 1, "name": "Ada", "admin": true },
      { "id": 2, "name": "Linus", "admin": false }
    ]
  },
  "meta": { "page": 1, "total": 2 }
}

export interface User {
  id: number;
  name: string;
  admin: boolean;
}

Path $.data.users[0]: this node only.

A response usually wraps the part you want in an envelope of paging and metadata. Path in Options takes a JSONPath, and the generator then reads only what it selects. A path to one place gives that value, and the root is named after the last name in the path; an index at the end names it after the singular of the list above, which is how users[0] becomes User.

A path with a filter or a wildcard can match many values, so it always gives an array of them, and the types describe a list. The line under the code says which of the two you got. Convert one part of a document explains that rule in full, since Convert’s Path is the same field. In the editor, the Tree view offers Generate code from this node on every object and array row, which opens the panel on that branch by pointing instead of typing.

A JSON Schema, and checking the next file

A JSON Schema is the one target that can check data rather than just describe it. A JSON Schema, and checking the next file generates one from a product record, then holds the next record to it in the editor’s Schema mode, where a wrong type and a missing field are both reported with their paths.

Structured output for OpenAI, Claude and Gemini

Asking a model to answer as JSON works better with a schema to hold it to, and each provider accepts a slightly different dialect. Structured output for OpenAI, Claude and Gemini writes the three for one list of tasks, and reads the report under each that says what the provider needed changed.