draccus

What it is and what it does

Draccus is a configuration framework that binds command-line arguments and YAML/TOML/JSON config files directly to Python dataclasses. It extends Pyrallis with support for config file inclusion (allowing you to split configs across files), choice registries (runtime selection between different config subtypes), and better handling of nested config containers. Your dataclass fields become both type-checked configuration parameters and IDE-autocompleted attributes; the framework parses arguments from the command line, a config file, or both, and initializes your dataclass with the result.

The library depends on pyyaml, toml, mergedeep, and typing-inspect to handle parsing, merging, and type resolution. It's designed for projects that need structured, nested configurations—particularly machine learning workflows where you might choose between different model architectures or optimizer types at runtime. The decorator-based API (@draccus.wrap) keeps the boilerplate minimal.

Use it for:

• Machine learning experiment configs: define model, optimizer, and training parameters as nested dataclasses and switch between them via config files.
• Multi-environment deployment: split base config, environment-specific overrides, and secrets into separate files and include them in a main config.
• CLI tools with complex options: replace argparse boilerplate with a clean dataclass definition and automatic --help generation.
• Configuration inheritance: use choice registries to let users select between different implementations (e.g., database backends) at startup.
• Config validation and IDE support: leverage dataclass type hints for autocomplete and static type checking of your configuration structure.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

Draccus parses command-line arguments and configuration files into Python dataclasses, with support for nested configs, config file inclusion, and runtime type selection via choice registries.

Yes. Draccus is actively maintained, has no known vulnerabilities, and solves a real problem—structured config management with IDE support—more cleanly than raw argparse or manual YAML loading. The low install friction and permissive license make it a low-risk addition. The Alpha status reflects its origin as a fork, not instability; it's already in production use. Install it if you need nested configs, config file inclusion, or runtime type selection; skip it if your config needs are simple enough for a basic argparse setup.

Install

draccus on PyPI

pip install draccus

uv add draccus

poetry add draccus

Installing draccus

Before you install

Low install friction with a pure-Python wheel and four lightweight runtime dependencies. Actively maintained with a recent release; marked as Alpha but in use with steady download volume.

License in practice

MIT license permits unrestricted use, modification, and distribution in both open-source and proprietary projects.

Quickstart

from dataclasses import dataclass
import draccus

@dataclass
class TrainConfig:
    workers: int = 8
    exp_name: str = 'default_exp'

@draccus.wrap()
def main(cfg: TrainConfig):
    print(f"Training {cfg.exp_name} with {cfg.workers} workers...")

if __name__ == "__main__":
    main()

Verify before relying

• Whether config file inclusion via !include tag works with all YAML edge cases or only standard structures
• Performance characteristics when handling deeply nested configs or large config files
• Compatibility with Python 3.11+ beyond the stated 3.9 and 3.10 support

Package facts

Evidence: draccus-0.11.6-py3-none-any.whl

Tags

More Application Frameworks packages