Redactix

Redactix is a lightweight Rust-backed Python library for detecting and redacting common PII in text.

Redactix is currently an alpha release. The detector set is intentionally small, and public APIs may change before a stable 1.0 release.

• Rust-backed speed for high-throughput text redaction.
• Simple Python API for detection, redaction, and reporting.
• Lightweight design focused on common structured PII and custom regex detectors.

Default Detectors

Name	Detection method	Validator
email	Regex	NULL
phone	Regex	Boundary checks; 10-15 digits; rejects repeated single-digit numbers
credit_card	Regex	Luhn checksum; 13-19 digits

Comparison

Redactix is focused on lightweight, deterministic text redaction for common structured PII. It is intentionally smaller than full PII detection frameworks.

Library	Difference
Redactix	Lightweight Rust-backed Python library with a simple API, built-in regex/validator detectors, custom regex detectors, and deterministic overlap resolution.
Microsoft Presidio	Much broader framework: NLP, regex, checksums, context-aware recognizers, multilingual support, text/images/structured data, Docker/PySpark/K8s options. More powerful but heavier. Source: Presidio README.
scrubadub	Python free-text scrubber with more built-in PII types: names, addresses, DOBs, URLs, credentials, SSNs, tax IDs, etc., plus optional/external detectors. Source: scrubadub docs.

Installation

Install from PyPI:

pip install pyredactix

Import the package as:

import redactix

For local development builds:

uv run maturin develop

Quick Start

import redactix

redactor = redactix.Redactor(
    detectors=["email", "credit_card"],
    mask_strategy="placeholder",
)

redactor.register_detector(
    name="employee_id",
    pattern=r"\bEMP\d{6}\b",
    placeholder="{{EMPLOYEE_ID}}",
    enabled=True,
)

cleaned = redactor.redact(text)
detections = redactor.detect(text)
report = redactor.redact_with_report(text)

redact() returns only redacted text:

redactix.redact("Email alex@example.com")
# "Email {{EMAIL}}"

detect() returns structured Detection dataclasses:

[
    redactix.Detection(
        start=6,
        end=22,
        value="alex@example.com",
        entity_type="EMAIL",
        detector_name="email",
        replacement="{{EMAIL}}",
    )
]

redact_with_report() returns a RedactionResult dataclass with both the cleaned text and detections used for redaction.

Redactor API

redactix.Redactor(
    detectors: Optional[Sequence[str]] = None,
    mask_strategy: "placeholder" | "fixed" | "length_preserving" = "placeholder",
    placeholder_format: str = "{{{entity_type}}}",
    mask_char: str = "*",
    fixed_mask: str = "***",
)

detectors=None enables all built-in detectors. Pass a subset such as ["email"] to enable only those detectors, or [] to start with no built-ins and register only custom detectors.

Masking Strategies

• placeholder replaces matches with stable placeholders such as {{EMAIL}}.
• fixed replaces every match with fixed_mask, which defaults to ***.
• length_preserving replaces each Python character in the match with mask_char.

Examples:

redactix.redact("Email alex@example.com", mask_strategy="fixed")
# "Email ***"

redactix.redact("Email alex@example.com", mask_strategy="length_preserving")
# "Email ****************"

Custom Detectors

Register custom regex-based detectors on a Redactor:

redactor = redactix.Redactor(detectors=[])
redactor.register_detector(
    name="employee_id",
    pattern=r"\bEMP\d{6}\b",
    placeholder="{{EMPLOYEE_ID}}",
    enabled=True,
    priority=100,
)

Detector names are normalized to lowercase identifiers for detector_name; placeholders and entity types use uppercase forms such as EMPLOYEE_ID.

Overlapping matches are resolved deterministically:

• Higher priority wins.
• If priority is equal, the longer match wins.
• Remaining ties are resolved by source position and registration order.

Roadmap

• Add more common PII detectors and localized PII patterns.
• Add optional AI-powered NER detection for advanced use cases while keeping the core library lightweight.

Contributing

Development setup, test commands, and contribution guidance are documented in CONTRIBUTING.md.

License

Redactix is distributed under the MIT License. See LICENSE for details.