Show HN: Open-Source DLP for LLMs (github.com)

A new open-source tool, ceil-dlp, has been launched as a Data Loss Prevention (DLP) plugin for LiteLLM, designed to automatically detect and safeguard Personally Identifiable Information (PII) in various formats, including text, images, and PDFs. The plugin blocks, masks, or logs sensitive data before it reaches LLM providers, thereby mitigating the risk of data leaks involving sensitive information like API keys. With growing concerns around data privacy regulations such as HIPAA, GDPR, and CCPA, ceil-dlp addresses a critical need for stronger compliance measures in AI deployments. Technically, ceil-dlp boasts a sophisticated ensemble detection approach, integrating multiple models to enhance PII detection accuracy while reducing false positives. Users can configure the DLP settings for different models through regex patterns within a single policy file, offering flexibility in guarding data based on model-specific needs. Additionally, it employs multi-pass OCR techniques for images and PDFs, ensuring comprehensive PII detection. This DLP solution consolidates functionalities that are typically fragmented across various tools, making it a versatile option for organizations handling sensitive data in AI environments. With an accessible installation process and support for community contributions, ceil-dlp aims to empower teams working with LLMs to maintain robust data protection practices efficiently.