/tools/json-to-python
JSON to Python Converter
Generate Python 3.10+ Pydantic v2 models, keyword-only dataclasses or precise TypedDict definitions.
JSON sample
Python output
/usage
Generate Python models from real JSON
- Paste a representative API response or choose a JSON file.
- Select Pydantic v2, dataclasses or TypedDict output and name the root model.
- Generate, review uncertain unions or empty arrays, then copy or download the Python file.
/inference
Python output that preserves the evidence
Pydantic v2 by default
Renamed fields use aliases and ConfigDict(populate_by_name=True), so awkward JSON keys remain serialisation-safe.
Missing differs from null
Fields missing from some array objects become optional. Explicit JSON nulls remain visible in the generated type.
Alternative standard-library shapes
Keyword-only dataclasses and TypedDict output support projects that do not use Pydantic.
Bounded local work
The shared inference worker records shape evidence rather than duplicate values, with depth, traversal, schema and output limits.
/choose output
Pydantic, dataclass or TypedDict?
| Python output | Best fit | Generated behaviour | Representative case |
|---|---|---|---|
| Pydantic v2 | Runtime validation and API boundaries | BaseModel, field aliases and populate_by_name | |
| Dataclasses | Standard-library data containers | Keyword-only classes with safe field ordering | |
| TypedDict | Static typing for dictionary-shaped data | Required and NotRequired keys without runtime conversion |
Use more than one representative object when fields vary. Empty or mixed arrays correctly produce uncertainty that should be reviewed rather than guessed. The public inference fixtures and reproduction notes cover nested, optional, aliased and uncertain shapes.
/faq
JSON to Python questions
Does it support nested arrays and objects?
Yes. Every sampled array item contributes evidence to deterministic nested models.
Why does an empty array use Any?
An empty array contains no element evidence. Add a representative item or replace Any with the domain type after generation.
Is generated code a finished domain model?
It is a strong starting contract from observed data. Review semantic constraints, validators and domain naming before production use.