I built an open-source RAG app for searching the Epstein files (36,000+ documents).
Every answer links to the exact source chunk. Click to verify. No black-box AI.
Try it at https://epfiles.ai (10 free messages, then use your own xAI key)
Stack: Next.js + FastAPI + ChromaDB + xAI (Grok).
What's included:
- Pre-built ChromaDB (auto-downloads on first run)
- Document chunks (~190MB, separate download)
- Scripts to regenerate embeddings with your own model
- One-command Docker setup
Limitations:
- LLM can still hallucinate, citations let you verify fast
- Fixed corpus (House Oversight release only)
- Requires API keys: OpenAI (embeddings), xAI (generation)
GitHub: https://github.com/benbaessler/epfiles
