wardcat: Open-Source On-Premise Data Privacy Tool for LLM and RAG Pipelines

A developer has released wardcat, an open-source Python library on PyPI designed to prevent sensitive data leaks when using large language models and RAG pipelines entirely on-premise. The tool uses a hybrid layered architecture combining Regex, SpaCy NER, and local open-weights LLMs such as Qwen3:14b to detect and anonymize sensitive information with high accuracy. A key feature called reversible masking allows anonymized text to be restored to its original form after LLM processing, while an is_sensitive() function filters prompts through a contextual security check before they are handled. Benchmarks comparing wardcat against Microsoft Presidio show wardcat's LLM layer achieving an F1 score of 0.920 on challenging scenarios and 88% accuracy in sensitivity classification, outperforming Presidio across all tested datasets. The project is publicly available on GitHub and PyPI, and the developer is welcoming contributions and feedback from the community.