filetype-detector

A Python library for detecting file types using multiple inference strategies, including path-based extraction, magic number detection, and AI-powered content analysis.

Accuracy

Metric	Value
Magic (libmagic) correct	59.9%
Magika (Google AI) correct	62.5%
Hybrid (runtime strategy)	74.2%
Total file formats	598

Generated from 598 canonical fixtures. See Accuracy Benchmarks for per-format data.

Features

Multiple Inference Methods: Choose from lexical, magic-based, AI-powered, or hybrid inference strategies
Type-Safe API: Type hints and type-safe inference method selection
Flexible Input: Supports both Path objects and string paths
Performance Optimized: Hybrid inference combines Magic and Magika when it improves the result
Well-Tested: Comprehensive automated test suite
Extensible: Base class architecture for custom inferencer implementations

Installation

Python Package

pip install filetype-detector

Or using rye:

rye sync

Interactive Terminal Demo

Use the installed terminal interface to filter a directory live and compare Lexical, Magic, Magika, and Hybrid results for the selected file:

filetype-detector-demo path/to/files

python -m filetype_detector path/to/files is equivalent. The directory is optional and defaults to the current working directory.

System Dependencies

Important: MagicInferencer and HybridInferencer require the libmagic system library to be installed.

Ubuntu/Debian

sudo apt-get update
sudo apt-get install libmagic1

Fedora/RHEL/CentOS

sudo dnf install file-libs
# or for older versions:
# sudo yum install file-libs

Arch Linux

sudo pacman -S file

macOS

Using Homebrew:

brew install libmagic

Using MacPorts:

sudo port install file

Windows

Windows users need to use python-magic-bin as an alternative:

pip install python-magic-bin

Or download libmagic DLL manually from file.exe releases.

Alpine Linux (Docker)

apk add --no-cache file

Verification

After installation, verify libmagic is available:

file --version

If the command works, libmagic is properly installed.

Quick Start

Recommended: Use AutoInferencer with backend="hybrid" for the best balance of performance and accuracy:

Content-based backends require the supplied path to reference an existing file.

from filetype_detector import AutoInferencer

inferencer = AutoInferencer(backend="hybrid")
ft = inferencer.infer("document.pdf")
'.pdf' in ft.extensions   # True
ft.mime_types             # ('application/pdf',)

All infer() methods return a FileType instance. See FileType for details.

For more examples and usage patterns, see the documentation site.

FileType

infer()가 반환하는 FileType은 frozen dataclass로, 확장자와 MIME 타입을 함께 보관한다.

from filetype_detector import FileType

ft = FileType.from_extension('.pdf')
ft.extensions   # ('.pdf',)
ft.mime_types   # ('application/pdf',)

ft2 = FileType.from_mimetype('image/jpeg')
ft2.extensions  # ('.jfif', '.jpe', '.jpeg', '.jpg') – OS별로 다를 수 있음

프로퍼티	타입	설명
`extensions`	`tuple[str, ...]`	연관된 파일 확장자 목록
`mime_types`	`tuple[str, ...]`	연관된 MIME 타입 목록

팩토리 메서드: FileType.from_extension(ext), FileType.from_mimetype(mime)

Performance Comparison

Choose the right inferencer based on your needs:

Inferencer	Avg. Time (per file)	Memory	Throughput	Best For
LexicalInferencer	< 0.001ms	Minimal	50,000+ files/sec	Trusted extensions
MagicInferencer	~1-5ms	Low	200-500 files/sec	Content-based detection
MagikaInferencer	~5-10ms*	High**	100-200 files/sec	Highest accuracy (text)
HybridInferencer	~1-6ms	Medium	150-400 files/sec	⭐ Recommended default

* After initial model load (~100-200ms one-time overhead)
** Model loaded into memory (~50-100MB)

Recommendation

For most use cases: Use AutoInferencer(backend="hybrid") - it delegates to HybridInferencer, which automatically uses Magic for binary files and Magika for text files.

See Choosing an Inferencer for a full decision guide and known limitations per inferencer.

Available Inferencers

LexicalInferencer

Fastest method - extracts file extensions directly from paths without reading file contents.

from filetype_detector import LexicalInferencer

inferencer = LexicalInferencer()
ft = inferencer.infer("document.pdf")
'.pdf' in ft.extensions   # True

inferencer.infer("file_without_ext")  # Raises ValueError

MagicInferencer

Uses python-magic (libmagic) to detect file types based on magic numbers and file signatures.

from filetype_detector import MagicInferencer

inferencer = MagicInferencer()
ft = inferencer.infer("file.dat")   # reads file content, ignores extension
ft.extensions                        # e.g. ('.pdf',) even if the name says .dat
ft.mime_types                        # e.g. ('application/pdf',)

System Requirements: Requires libmagic system library. See Installation section.

MagikaInferencer

AI-powered detection with confidence scores. Especially effective for text files.

from filetype_detector import MagikaInferencer

inferencer = MagikaInferencer()
ft = inferencer.infer("script.py")
'.py' in ft.extensions   # True

# infer_with_score returns (str, float), not FileType
extension, score = inferencer.infer_with_score("data.json")  # ('.json', 0.98)

HybridInferencer

Smart two-stage approach: uses Magic for all files, then Magika for text files.

from filetype_detector import HybridInferencer

inferencer = HybridInferencer()

# Text file: Magic detects text/plain, then Magika refines to .py
ft = inferencer.infer("script.py")
'.py' in ft.extensions   # True

# Binary file: Magic result is used directly, Magika is skipped
ft = inferencer.infer("document.pdf")
'.pdf' in ft.extensions  # True

If you want the same behavior through the unified interface, use AutoInferencer(backend="hybrid").

System Requirements: Requires libmagic system library. See Installation section.

Choosing an Inferencer

파일 성격과 요구사항에 따라 추론기를 선택한다.

파일 이름(확장자)을 신뢰할 수 있는가?
├─ 예  → LexicalInferencer   (파일 읽기 없음, 가장 빠름)
└─ 아니오 / 불확실
       ├─ 주로 바이너리? (PDF, Office, 이미지, 압축 등)
       │       └─ MagicInferencer
       ├─ 주로 텍스트? (Python, JSON, HTML, CSV 등)
       │       └─ MagikaInferencer   (세부 분류 + 신뢰도 점수)
       └─ 혼합 / 미확정
               └─ HybridInferencer  (기본값 추천)

상황	추천
CI 파이프라인 등 이미 정제된 파일	`LexicalInferencer`
확장자 없는 파일, 잘못된 확장자 검증	`MagicInferencer`
`.py` / `.json` / `.csv` 등 텍스트 세분화	`MagikaInferencer`
신뢰도 점수(`infer_with_score`)가 필요할 때	`MagikaInferencer`
업로드 파일 등 무엇이 올지 모를 때	`HybridInferencer`
단일 진입점으로 backend만 교체하고 싶을 때	`AutoInferencer`

Known Limitations

LexicalInferencer

확장자가 없거나 잘못된 경우 탐지 불가. 파일 내용을 읽지 않으므로 위변조 파일을 그대로 통과시킨다.

MagicInferencer

text/plain으로 분류되는 파일들의 세부 구분이 불가능하다. Python, JSON, CSV 등이 동일하게 text/plain을 반환한다. Compound Document 포맷(구 .doc, .ppt, .xls)은 같은 컨테이너를 공유하므로 여러 확장자가 후보로 반환된다.

MagikaInferencer

텍스트 파일 외에는 이점이 제한적이다. 바이너리 파일에서는 MagicInferencer와 동등하거나 열등한 경우가 있다.
HWP: 학습 데이터에 없어 빈 결과를 반환한다.
ZIP 기반 포맷 (HWPX, ODF, ePub 등): 내부 구조를 분석하지 않아 .epub 등으로 오분류될 수 있다.
모델 로딩 비용: 첫 인스턴스화 시 100–200ms의 오버헤드가 있다.

HybridInferencer

Magic 결과가 text/로 시작할 때만 Magika를 호출한다. ZIP 기반 포맷(HWPX, ODF 등)과 Compound Document 포맷은 Magic으로만 처리되며 Magika가 적용되지 않는다.

Key Features

✅ Multiple inference strategies - Choose the right method for your use case
✅ Type-safe API - Full type hints and type-safe method selection
✅ Flexible input - Supports both Path objects and string paths
✅ Performance optimized - Hybrid inference combines fast binary detection with more detailed text detection
✅ Well-tested - Comprehensive test suite
✅ Extensible - Base class architecture for custom implementations

Documentation

Tutorial: Getting Started
How-to Guides: choosing an inferencer, bulk processing, and extending the library
Reference: API pages for AutoInferencer, BaseInferencer, and each inferencer
Explanation: inference strategy trade-offs and architecture notes

For detailed usage examples, error handling, and advanced patterns, see the documentation site.

Testing

Truth data

Canonical fixture truth lives in tests/truth/canonical_fixtures.json. Tool snapshots (version-pinned per-inferencer results) live in tests/truth/tool_snapshots.json.

Regenerate truth data after adding fixtures:

# Rebuild tool snapshots from canonical fixtures
rye run python scripts/generate_truth_data.py

# Check that existing snapshots match without overwriting
rye run python scripts/generate_truth_data.py --check

Accuracy suites

Cross-suite smoke test (representative sample across all 598 canonical fixtures):

rye run python -m pytest tests/test_file_type_truth.py tests/test_fixture_coverage.py -v

Per-inferencer accuracy (semantic + snapshot contracts):

rye run python -m pytest tests/test_accuracy_magic.py tests/test_accuracy_magika.py tests/test_accuracy_hybrid.py -v

Delegation correctness (AutoInferencer matches direct inferencer):

rye run python -m pytest tests/test_auto_inferencer.py -v

Behavior tests (control flow, mocks, error handling):

rye run python -m pytest tests/test_magic_inferencer.py tests/test_magika_inferencer.py tests/test_hybrid_inferencer.py -v

Full verification

One command to run the complete verification pipeline — truth check, all tests, lint, and type-check:

rye sync && rye run python scripts/generate_truth_data.py --check && rye run python -m pytest && rye run ruff check src tests scripts examples && rye run mypy src/filetype_detector/ --ignore-missing-imports

Architecture

Base Class

All inferencers inherit from BaseInferencer, which defines a common interface:

from abc import ABC, abstractmethod
from typing import Union
from pathlib import Path
from filetype_detector import FileType

class BaseInferencer(ABC):
    @abstractmethod
    def infer(self, file_path: Union[Path, str]) -> FileType:
        """Infer file type from path. Returns a FileType instance."""
        raise NotImplementedError

Custom Inferencer

You can create custom inferencers by subclassing BaseInferencer:

from filetype_detector import BaseInferencer, FileType
from typing import Union
from pathlib import Path

class CustomInferencer(BaseInferencer):
    def infer(self, file_path: Union[Path, str]) -> FileType:
        # Your custom logic here
        return FileType.from_extension(".custom")

Dependencies

Runtime dependencies:

python-magic>=0.4.27: Magic-number file detection
magika>=1.0.1: AI-powered content detection
rich>=14.2.0: Styled inference output
textual>=8.2.8: Interactive terminal interface

Development, documentation, and fixture-generation packages are isolated in the dev, docs, and fixtures optional extras.

Requirements

Python >= 3.10

License

This project is open source. See LICENSE file for details.

Contributing

Contributions are welcome! Please ensure:

All tests pass: pytest tests/ -v
Code follows the existing style
New features include appropriate tests
Documentation is updated

Acknowledgments

python-magic for libmagic bindings
Google Magika for AI-powered file type detection

Name		Name	Last commit message	Last commit date
Latest commit History 57 Commits
.github/workflows		.github/workflows
docs		docs
examples		examples
scripts		scripts
src/filetype_detector		src/filetype_detector
tests		tests
.gitignore		.gitignore
.python-version		.python-version
LICENSE		LICENSE
README.md		README.md
mkdocs.yml		mkdocs.yml
pyproject.toml		pyproject.toml
pytest.ini		pytest.ini
requirements-dev.lock		requirements-dev.lock
requirements.lock		requirements.lock

Uh oh!

Folders and files

Latest commit

History

Repository files navigation

filetype-detector

Accuracy

Features

Installation

Python Package

Interactive Terminal Demo

System Dependencies

Ubuntu/Debian

Fedora/RHEL/CentOS

Arch Linux

macOS

Windows

Alpine Linux (Docker)

Verification

Quick Start

FileType

Performance Comparison

Recommendation

Available Inferencers

LexicalInferencer

MagicInferencer

MagikaInferencer

HybridInferencer

Choosing an Inferencer

Known Limitations

LexicalInferencer

MagicInferencer

MagikaInferencer

HybridInferencer

Key Features

Documentation

Testing

Truth data

Accuracy suites

Full verification

Architecture

Base Class

Custom Inferencer

Dependencies

Requirements

License

Contributing

Acknowledgments

About

Resources

License

Contributing

Uh oh!

Stars

Watchers

Forks

Releases 1

Packages 0

Uh oh!

Uh oh!

Contributors

Uh oh!

Languages

Packages