Generate Python TypedDicts from Complete JSON Examples

Complete documents and all array elements

Paste strict JSON or choose a local UTF-8 file. Generate explicitly from one document or up to twenty complete examples of the same contract. Every array element and document contributes, including fields introduced at the end. A root array remains an array. Empty examples, decoded duplicate keys, comments, trailing commas and concatenated documents stop generation and preserve the input.

The editor shows the active example and total count. After generating, review the filename, root type expression, JSON node count and size of the complete code. These details describe the current result; copying and downloading include the full file even when the preview is shortened.

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

from typing import NotRequired, TypeAlias, TypedDict

RootItem = TypedDict("RootItem", {
  "id": int,
  "nombre": NotRequired[str],
})

Root: TypeAlias = list[RootItem]

Presence, None and exact keys

NotRequired[T] means a key was absent from some objects in its context; T | None means an observed null. For [{"x":null},{"x":"a"},{}], the field is NotRequired[str | None]. A required null-only field uses None, an optional null-only field uses NotRequired[None]. Absence does not add None or defaults. Objects in different fields have separate contracts.

Functional TypedDict syntax preserves dictionary keys such as class, a-b, the empty string, quotes, emoji and isolated surrogate escapes; only code identifiers are adjusted for collisions. Dependencies are declared before use. Python 3.11+ and only its standard library are required. TypedDict describes ordinary dictionaries and does not validate JSON or enforce required keys at runtime; no deserializer, dataclass, Pydantic model, dates or enums are generated.

Root [1,"x",null] becomes list[int | str | None]; [null] uses list[None], [] uses list[object] with a notice, and {} uses dict[str, object] because no fields were observed. A root null alias uses types.NoneType. Root primitive and union aliases use TypeAlias.

Numbers and complete local exports

Pure integer lexemes use int, including integers outside JavaScript’s safe range. A point or exponent selects float, also for mixed integer/decimal evidence. A parser precision notice remains visible even if int can preserve the original number; float emits an approximation notice when precision evidence is lost. Integer -0 produces a notice because int normalizes its sign. No sample values or coercions appear in the output.

Limits: 20 samples, 1,000,000 UTF-16 input units, 4,000,000 UTF-8 bytes in total and per file, depth 200, 100,000 nodes, 1,000,000 parse work units, 100,000 inference visits, 10,000 types, 20,000 fields, 2,000,000 merge work units, 2,000 declarations and 4,000,000 emission work units. Output allows 2,000,000 UTF-16 units and 8,000,000 bytes. Limits cannot necessarily be reached together; the operation times out after 15 seconds with no partial code.

The preview contains at most 20,000 UTF-16 units; copy and download include the entire current types.py, UTF-8 without BOM, LF and a final newline, as text/plain;charset=utf-8. At most 100 diagnostic details and 50 name rows totaling 10,000 units are shown, with omitted counts. Denied clipboard access offers explicit manual copying of the complete code. One initial BOM is removed for parsing; error positions refer to the visible original document. Cancel terminates the Worker and keeps all examples. Edits, options, sample order and mode changes invalidate output and exports.

Processing stays in a local browser Worker without uploads, storage or data in navigation links. JavaScript and Worker support are required. The page never imports, compiles or executes generated Python.