pdf-anonymizer-core 0.3.3

A core library to anonymize PDF, Markdown, and plain text files using LLMs.

🦉🫥 PDF Anonymizer Core

This package provides the core functionality for the PDF/Text anonymizer, including text extraction, LLM-driven anonymization, and deanonymization logic. It is used by pdf-anonymizer-cli.

Installation

Install the base package with your favorite package manager:

pip install pdf-anonymizer-core

To use a specific LLM provider, you must install the corresponding extra. This helps to keep the installation lightweight by only downloading the libraries you need.

• Google: pip install "pdf-anonymizer-core[google]"
• Ollama: pip install "pdf-anonymizer-core[ollama]"
• Hugging Face: pip install "pdf-anonymizer-core[huggingface]"
• OpenRouter: pip install "pdf-anonymizer-core[openrouter]"
• OpenAI: pip install "pdf-anonymizer-core[openai]"
• Anthropic: pip install "pdf-anonymizer-core[anthropic]"

You can also install multiple extras at once:

pip install "pdf-anonymizer-core[google,ollama]"

Full documentation, including the Python SDK guide, CLI reference, recipes, and architecture, lives at leo-gan.github.io/anonymizer/ (especially the Recipes & Common Workflows section).

Environment Variables

The core library itself does not load .env files. Environment variables must be loaded by the application that uses this library (e.g., pdf-anonymizer-cli) or set in your shell.

• GOOGLE_API_KEY: Required when using Google models.
• HUGGING_FACE_TOKEN: Required when using Hugging Face models.
• OPENROUTER_API_KEY: Required when using OpenRouter models.
• OPENAI_API_KEY: Required when using OpenAI models.
• ANTHROPIC_API_KEY: Required when using Anthropic models.
• OLLAMA_HOST: Optional, defaults to http://localhost:11434 when using Ollama models.

API Usage

anonymize_file()

Anonymizes a single file and returns the anonymized text and a mapping of original entities to their placeholders.

from pdf_anonymizer_core.core import anonymize_file
from pdf_anonymizer_core.prompts import detailed

# Example of programmatic usage
text, mapping = anonymize_file(
    file_path="/path/to/file.pdf",
    prompt_template=detailed.prompt_template,
    model_name="gemini-2.5-pro"  # Can also be a new model like "google/gemini-flash-latest"
)

if text and mapping:
    print("Anonymized Text:", text)
    print("Mapping:", mapping)

deanonymize_file()

Reverts anonymization using a mapping file. The function writes the restored document and a stats file to conventional locations and returns their paths.

from pdf_anonymizer_core.utils import deanonymize_file

# Assumes you have an anonymized file and a mapping file
deanonymized_file_path, stats_file_path = deanonymize_file(
    "path/to/anonymized.md",
    "path/to/mapping.json",
)

print("Deanonymized file:", deanonymized_file_path)
print("Stats file:", stats_file_path)

Configuration

You can import default configurations and available models from the conf module.

from pdf_anonymizer_core.conf import (
    DEFAULT_MODEL_NAME,
    ModelName,
    PromptEnum,
)

print(f"Default model: {DEFAULT_MODEL_NAME}")
print(f"Available Google models: {[m.value for m in ModelName if m.provider == 'google']}")

---

See Also

• Main Documentation — Full project guides and 101 course.
• CLI Package README — Command line interface usage.
• Recipes & Common Workflows — Practical SDK examples (including profiles, caching, and advanced usage).
• API Reference (auto) — Detailed function signatures.
• Troubleshooting — Common issues when using the core library.
• Architecture — How the anonymization pipeline works internally.