epstein-search 0.1.1

Semantic search over the Epstein Files using zvec + sentence-transformers

epstein-search 🔍

Semantic search over the publicly available Epstein Files — court documents, FBI reports, and DOJ publications — using AI-powered vector search.

Built with zvec and pre-computed embeddings from devankit7873/EpsteinFiles-Vector-Embeddings-ChromaDB.

No API key needed for search. Runs entirely locally.

---

☕ Support This Project

This is free, open-source research tooling. If you find it useful, please consider supporting:

Or tip via crypto:

Chain	Address
SOL	76fCU6va3cGrbak4i9mwFfdYx1QsJJrqrxViFDUTsXUL
ETH	0xE808754a18A893A3eeFE01780D822C902680d1B7
BASE	0xE808754a18A893A3eeFE01780D822C902680d1B7

---

Quick Start

pip install epstein-search

# One-time setup — downloads pre-built index (~100K+ document chunks)
epstein-search setup

# Search — free, local, no API key needed
epstein-search search "flight logs to the island"

# Filter by document type
epstein-search search "testimony" --doc-type deposition

# AI-powered answers (local Ollama = free, or any LLM API)
epstein-search ask "Who appears most in the flight logs?" --model ollama/llama3

Commands

Command	Description	API Key?
epstein-search setup	Download & build search index	No
epstein-search chat	Interactive mode (recommended)	Only for cloud LLMs
epstein-search search "query"	One-off semantic search	No
epstein-search ask "question"	One-off RAG answer	Only for cloud LLMs
epstein-search info	Show index stats	No
epstein-search ingest	Build from scratch (advanced)	No

Interactive Mode (Recommended)

The easiest way to use epstein-search:

epstein-search chat

Type questions naturally. No flags needed. Commands inside chat:

Command	Description
/search	Switch to search-only mode (no LLM)
/ask	Switch to RAG mode (LLM answers)
/model openai/llama3	Change LLM on the fly
/topk 5	Change number of results
/info	Show current settings
/quit	Exit

Tip: Set your model once in .env so you never need to specify it:

EPSTEIN_LLM_MODEL=openai/your-model-name

Search Options

epstein-search search "financial records" --top-k 5
epstein-search search "testimony" --source "FBI"
epstein-search search "flight logs" --doc-type flight_log
epstein-search search "court order" --json-output

Document Type Filters

Filter	Description
court_filing	Court motions, orders, filings
deposition	Sworn testimony, Q&A transcripts
fbi_report	FBI investigation reports
flight_log	Aircraft flight records
financial	Bank records, wire transfers
other	Miscellaneous documents

Ask (RAG)

The ask command retrieves relevant documents and generates an answer using any LLM via LiteLLM.

Free options (no API key):

# Run Ollama locally — completely free
epstein-search ask "Key individuals mentioned" --model ollama/llama3
epstein-search ask "What do the flight logs show?" --model ollama/mistral

LM Studio (local, no API key):

1. Download LM Studio, load a model, start the local server (default: port 1234)
2. Add to your .env file:

OPENAI_API_BASE=http://localhost:1234/v1
OPENAI_API_KEY=lm-studio
EPSTEIN_LLM_MODEL=openai/your-model-name

3. Then just run:

epstein-search chat   # interactive mode, no flags needed

Cloud LLM options (API key required):

# Gemini (requires GEMINI_API_KEY)
epstein-search ask "Summarize the findings" --model gemini/gemini-3-flash-preview

# OpenAI (requires OPENAI_API_KEY)
epstein-search ask "Timeline of events" --model gpt-4o

# Anthropic (requires ANTHROPIC_API_KEY)
epstein-search ask "Financial connections" --model anthropic/claude-sonnet-4-20250514

# Show source documents alongside the answer
epstein-search ask "Flight patterns" --show-sources

How It Works

pip install epstein-search
         ↓
epstein-search setup
  → Downloads 100K+ pre-computed embeddings (all-MiniLM-L6-v2)
  → Imports into local zvec vector database
         ↓
epstein-search search "your query"
  → Embeds query locally (sentence-transformers, no API key)
  → Vector similarity search via zvec
  → Returns matching document chunks

Dataset

Embeddings: devankit7873/EpsteinFiles-Vector-Embeddings-ChromaDB — 100K+ chunks, 384-dim vectors, based on the Epstein Files 20K corpus.

All content is from publicly available sources:

• U.S. Department of Justice Epstein Library
• House Oversight Committee releases
• Unsealed federal court documents
• FBI reports and DOJ publications

License

MIT — see LICENSE