ChatGPT for the sketchy parts of the internet

What it does Robin is a Streamlit app that routes dark-web search queries through Tor, then uses an LLM to refine the query, filter the results, and summarize findings into a readable investigation report. It saves sessions to disk for later review.

The interesting bit The architecture is deliberately modular: search, scrape, and LLM stages are separate, so you can swap in a local Ollama model or a remote Claude endpoint without touching the Tor plumbing. That matters because sending sensitive OSINT queries to third-party APIs is a compliance nightmare the README explicitly warns about.

Key highlights

• Supports OpenAI, Claude, Gemini, Ollama, Groq, LM Studio, or any OpenAI-compatible API
• Runs in Docker with Tor bundled; persists investigations across restarts via volume mount
• Streamlit UI with a “Past Investigations” sidebar for continuity
• Extensible: new search engines or output formats plug into the workflow
• Inspired by Thomas Roccia’s “Perplexity of the Dark Web” demo

Caveats

• Requires a running Tor instance; the tool does not handle Tor setup automatically
• LLM filtering happens server-side over third-party APIs, so sensitive queries leave your network unless you use a local model

Verdict Security researchers and law-adjacent investigators who need structured dark-web recon without manual Tor browsing will save time. Privacy purists running everything air-gapped will still need to wire up local LLMs carefully.