Migrate from Microsoft Presidio to Philter

Before and after: the same redaction in code

Because Presidio is a library, the most direct migration target is Phileas, the embeddable engine under Philter. Here is the same job (detect an email and an SSN, then de-identify both) in Presidio and in Phileas, side by side.

Presidio (Python)

In Presidio you call the analyzer, then build an operator config for the anonymizer:

from presidio_analyzer import AnalyzerEngine
from presidio_anonymizer import AnonymizerEngine
from presidio_anonymizer.entities import OperatorConfig

analyzer = AnalyzerEngine()
anonymizer = AnonymizerEngine()

text = "Contact john@example.com or call about SSN 123-45-6789."
results = analyzer.analyze(text=text, language="en")

anonymized = anonymizer.anonymize(
    text=text,
    analyzer_results=results,
    operators={
        "EMAIL_ADDRESS": OperatorConfig("replace", {"new_value": "<EMAIL>"}),
        "US_SSN": OperatorConfig("replace", {"new_value": "<SSN>"}),
    },
)
print(anonymized.text)
# Contact <EMAIL> or call about SSN <SSN>.

Phileas (Python)

In Phileas the same rules live in a policy, and the engine handles detection and manipulation in one call:

from phileas.policy.policy import Policy
from phileas.services.filter_service import FilterService

policy = Policy.from_dict({
    "name": "contact-redaction",
    "identifiers": {
        "emailAddress": {"emailAddressFilterStrategies": [
            {"strategy": "REDACT", "redactionFormat": "{{{REDACTED-%t}}}"}]},
        "ssn": {"ssnFilterStrategies": [
            {"strategy": "REDACT", "redactionFormat": "{{{REDACTED-%t}}}"}]},
    },
})

result = FilterService().filter(
    policy=policy,
    context="support",
    document_id="doc-001",
    text="Contact john@example.com or call about SSN 123-45-6789.",
)
print(result.filtered_text)
# Contact {{{REDACTED-email-address}}} or call about SSN {{{REDACTED-ssn}}}.

The same policy runs unchanged on the Java and .NET builds of Phileas: the policy schema is shared across all three runtimes.

The same policy in PhiSQL

Instead of hand-writing the policy JSON, you can express it in PhiSQL, which compiles to the same schema and is version-controlled and reviewable like any other code:

POLICY contact_redaction;

REDACT EMAIL_ADDRESS WITH REDACT(format='{{{REDACTED-%t}}}');
REDACT SSN          WITH REDACT(format='{{{REDACTED-%t}}}');

Translating the rest of your Presidio setup

The common Presidio building blocks map cleanly:

• Custom PatternRecognizer becomes a DEFINE IDENTIFIER statement (or a custom identifier entry in policy JSON). No Python class to subclass:

  DEFINE IDENTIFIER 'MRN' MATCHING '\bMRN[\s:#]*\d{5,}\b' CASE INSENSITIVE
    WITH REDACT(format='{{{REDACTED-MRN}}}');
  

• NLP-based recognizers (spaCy, Stanza, or transformer) become a DETECT PHEYE statement backed by PhEye PII/PHI-trained models:

  DETECT PHEYE LABELS ('PERSON') WITH REDACT;
  

• Anonymizer operators map to strategies: Presidio replace to a Phileas replace or static value, redact to REDACT, mask to MASK, hash to a hash strategy, encrypt to crypto-replace. Strategies Presidio does not have a built-in equivalent for include format-preserving encryption and date-shift:

  REDACT SSN  WITH FPE_ENCRYPT;
  REDACT DATE WITH SHIFT(days=30);
  

• Conditional logic that would be custom Python around Presidio becomes a WHERE predicate. For example, keep only the last four digits of a credit card when detection confidence is high:

  REDACT CREDIT_CARD WITH LAST_4 WHERE CONFIDENCE > 0.85;
  

A practical migration ports your recognizers and operators first to reach parity, then revisits the policy to use the conditional rules, consistent pseudonymization, and format-preserving strategies that were custom code (or simply unavailable) under Presidio.