Presidio Analyzer

Warnings

• breaking Presidio underwent a significant revamp from V1 to V2 (starting around 2.0.0). This involved a migration from gRPC to HTTP-based APIs, changes in JSON payload formats (structured objects to flattened JSON, camelCase to snake_case), and deprecation of some services. Code written for V1 is not compatible with V2 APIs.
  Fix: Refer to the 'Changes from V1 to V2' documentation on the Presidio website. Rewrite API calls and update data structures to conform to the new HTTP/JSON format.
• gotcha Many country-specific PII recognizers (e.g., for Singapore, Australia, Germany, Sweden) are disabled by default to prevent false positives when they are not explicitly needed. If you require detection for specific regional PII, you must explicitly enable these recognizers either via a YAML configuration file or programmatically by adding them to the RecognizerRegistry.
  Fix: Review the documentation for 'Recognizer registry from file' or programmatic `RecognizerRegistry` customization to enable specific country-specific recognizers.
• gotcha The AnalyzerEngine relies on NLP models (like spaCy) for many detections. While `spacy` itself is a dependency, the language models (e.g., `en_core_web_lg`) must be downloaded separately using `python -m spacy download <model_name>`. Failure to do so will result in errors or limited detection capabilities.
  Fix: Always run `python -m spacy download <model_name>` for the languages you intend to analyze, as part of your environment setup.
• gotcha Specific versions of underlying NLP libraries, particularly spaCy, might be explicitly restricted in certain `presidio-analyzer` releases. For example, `spacy.cli` was restricted for version 3.7.0 in release 2.2.356. Using an incompatible spaCy version can lead to unexpected behavior or errors.
  Fix: Always check the `presidio-analyzer` release notes and `pyproject.toml` for explicit dependency version constraints and ensure your environment adheres to them.
• gotcha Users employing static type checking (e.g., mypy) may encounter type errors in versions after 2.2.33, specifically related to the initialization of `AnonymizerEngine` and type mismatches for `RecognizerResult` between `presidio-analyzer` and `presidio-anonymizer` due to separate class definitions.
  Fix: Monitor GitHub issues for type-related fixes. Depending on the error, temporary workarounds might include `type: ignore` comments or casting until official fixes are released.

Install

• pip install presidio-analyzer Install core library
• python -m spacy download en_core_web_lg Download default spaCy NLP model

Imports

• AnalyzerEngine

  from presidio_analyzer import AnalyzerEngine

• RecognizerRegistry

  from presidio_analyzer.recognizer_registry import RecognizerRegistry

• NlpEngineProvider

  from presidio_analyzer.nlp_engine import NlpEngineProvider

Quickstart

This quickstart demonstrates how to initialize the AnalyzerEngine and detect PII entities in a given text. Ensure you have downloaded the `en_core_web_lg` spaCy model as instructed in the installation steps, as it's used by default for NLP capabilities. The output will show detected entities, their location, and a confidence score.

from presidio_analyzer import AnalyzerEngine

# Initialize the AnalyzerEngine
# This will load the default spaCy NLP model (en_core_web_lg if downloaded)
analyzer = AnalyzerEngine()

text = "My name is John Doe and my phone number is (123) 456-7890."

# Analyze the text for PII entities
# Specify entities to look for, or leave empty for all supported entities
results = analyzer.analyze(text=text, entities=["PERSON", "PHONE_NUMBER"], language='en')

for result in results:
    print(f"Entity: {result.entity_type}, Text: {text[result.start:result.end]}, Score: {result.score:.2f}")