py-data-cloak

py-data-cloak is a rule-based data anonymization engine for Python. It works on JSON, NDJSON, CSV, and SQL dumps from the command line, integrates with the Django ORM for safe dumpdata exports, and plugs into Pandas for data-science workflows.

It keeps referential integrity (user_id=42 always masks to the same value within a session — or across runs with a persistent vault), streams large files without loading them into memory, and can auto-detect PII from a sample to generate a starter rules file for you.

Install

pip install py-data-cloak

# Optional extras
pip install "py-data-cloak[django]"   # Django dumpdata_masked command + mixin
pip install "py-data-cloak[pandas]"   # mask_dataframe helper
pip install "py-data-cloak[vault]"    # encrypted persistent vaults
pip install "py-data-cloak[all]"      # all of the above

Quick start

# privacy_rules.yaml
email:           faker:email
ssn:             mask_all_but_last_4
password:        fixed:***REDACTED***
notes:           clear
credit_card:     format_preserve:luhn
salary:          noise:5000
dob:             shift_date:365
user_id:         hash:sha256:mysalt
'*_email':       faker:email           # glob: any column ending in _email
're:.*_secret$': redact                # regex
'user.profile.phone': format_preserve  # nested JSON path

pycloak process users.json   --rules privacy_rules.yaml --output safe.json
pycloak process huge.jsonl   --rules privacy_rules.yaml --output safe.jsonl
pycloak process export.csv   --rules privacy_rules.yaml --output safe.csv
pycloak process dump.sql     --rules privacy_rules.yaml --output safe.sql
pycloak process input.json   --rules privacy_rules.yaml --dry-run

Don't know what to mask? Let pycloak guess:

pycloak scan users.json --output starter_rules.yaml

Rules reference

Field matching

• Exact — "email", "user.profile.phone" (nested JSON path)
• Glob — "*_email", "user.*", "address.[lc]ine?"
• Regex — prefix the pattern with re:, e.g. "re:^.*_secret$"

Exact rules are checked first; glob and regex follow in declaration order, so put specific patterns above broader catch-alls.

Python API

from pycloak import Anonymizer, Vault, load_rules

rules = load_rules("privacy_rules.yaml")
vault = Vault("masked.vault", passphrase="hunter2")   # optional
anonymizer = Anonymizer(rules, seed=42, locale="en_US", vault=vault)

masked = anonymizer.process_record({"email": "alice@example.com", "salary": 95000})
anonymizer.save()   # persist the vault for the next run

Custom rules

from pycloak import register_rule
from pycloak.rules import Rule

@register_rule
class UppercaseRule(Rule):
    @classmethod
    def matches(cls, spec): return spec == "upper"
    @classmethod
    def parse(cls, spec):   return cls()
    def apply(self, value, ctx): return str(value).upper()

Or use custom:my.module:my_func directly in YAML:

# my_module.py
def my_func(value, ctx):
    return f"<masked:{ctx.field_name}>"

some_field: custom:my_module:my_func

Pandas

from pycloak.pandas_helper import mask_dataframe

masked_df = mask_dataframe(df, rules)

Django

Add to INSTALLED_APPS:

INSTALLED_APPS = [..., "pycloak.django"]
PY_DATA_CLOAK_MASKING_RULES = "privacy_rules.yaml"   # path or dict
PY_DATA_CLOAK_SAFE_REPR = True                       # optional: safer repr for logs

Stream-friendly masked dump:

python manage.py dumpdata_masked auth.User --output safe_users.json --chunk-size 5000

Per-instance access via the mixin:

from pycloak.django.mixins import MaskedModelMixin

class User(MaskedModelMixin, models.Model):
    ...

User.objects.first().masked_data()

Versioning

0.2.0 introduces streaming, consistency, SQL/NDJSON support, PII detection, vaults, Pandas/Django streaming, and the rule registry. The pre-1.0 API may still evolve; 1.0.0 will lock the public surface.

License

MIT.