has-anonymizer

On-device privacy masking anonymizes text and images locally before cloud processing. It supports multilingual text anonymization across 8 languages and image masking across 21 privacy categories while preserving document structure.

You can restore anonymized content using a mapping-based restore path. The tool generates per-file mappings during the hide phase, ensuring deterministic reconstitution of originally masked text and visual data after processing.

On-device image anonymization works without uploading data to external servers. It relies on a local llama-server runtime with auto-managed models to mask 21 privacy categories directly on your hardware.

You can batch scan documents for sensitive content before sharing using parallel processing and chunking features. The tool processes multiple files locally to identify and mask sensitive text and visual data across various privacy categories.

Limitations of using local models for privacy masking include dependency on a local llama-server runtime and specific Python libraries like ultralytics and opencv-python-headless. Your hardware resources directly dictate batch processing speed and parallelism.