Datafiles is a bidirectional serialization library for Python dataclasses to synchronizes objects to the filesystem using type annotations. It supports a variety of file formats with round-trip preservation of formatting and comments, where possible. Object changes are automatically saved to disk and only include the minimum data needed to restore each object.

Some common use cases include:

  • Coercing user-editable files into the proper Python types
  • Storing program configuration and state in version control
  • Loading data fixtures for demonstration or testing purposes
  • Synchronizing application state using file sharing services
  • Prototyping data models agnostic of persistence backends


Install it directly into an activated virtual environment:

$ pip install datafiles

or add it to your Poetry project:

$ poetry add datafiles

Quick Start

Decorate a type-annotated class with a directory pattern to synchronize instances:

from datafiles import datafile

class Sample:

    key: int
    name: str
    value: float = 0.0

By default, all member variables will be included in the serialized file except for those:

  • Included in the directory pattern
  • Set to default values

So, the following instantiation:

>>> sample = Sample(42, "Widget")

produces samples/42.yml containing:

name: Widget

and the following instantiation restores the object:

>>> from datafiles import Missing
>>> sample = Sample(42, Missing)

Type Checking

If using mypy, enable the plugin via pyproject.toml settings:


plugins = "datafiles.plugins:mypy"

or mypy.ini configuration file:


plugins = datafiles.plugins:mypy