daxa-ai/pebblo

Pebblo uses semantic and entity classification to automatically discover PII, sensitive data, and topics within loaded documents, with results surfaced through a web UI or PDF reports. It provides three integrated components: a REST server for data inspection, SafeLoader wrappers for LangChain document loaders that intercept data pre-ingestion, and SafeRetriever chains that enforce identity-based and semantic access controls on vector database queries before LLM inference.

149 stars. No commits in the last 6 months. Available on PyPI.