Epstein Files RAG

Epstein Files RAG is an open-source AI document-search tool that lets users explore unsealed Jeffrey Epstein court documents through conversational, retrieval-augmented search. It is aimed at researchers, journalists, and analysts who want to query investigative records in a searchable, chat-style interface instead of reading raw filings. The project is fully open-source with no proprietary dependencies and can run locally via Ollama or against cloud inference through Groq or OpenRouter. An automated ingestion step downloads and indexes a curated dataset from Hugging Face (the Nikity/Epstein-Files collection in Apache Parquet format), and strict guardrails keep responses within the investigative-document context rather than answering open-ended questions. The stack is LangChain for orchestration, ChromaDB as the vector database, Streamlit for the interface, and Python 3.10 or higher. Setup requires Python 3.10+, an optional Ollama install for local models, and API keys for Groq or OpenRouter if using cloud inference; the first data chunk ingests in roughly 3 to 5 minutes at about 0.5 GB by default, and a virtual environment is recommended. As an open-source repository, there is no pricing; running it costs only any inference or hosting you choose.

Researchers and journalists use Epstein Files RAG to query unsealed Epstein court documents conversationally, running it locally with Ollama or via cloud LLMs.