I Gave Pi-hole an AI Brain (And Now It Learns From My Decisions)

🛑 Stop Guessing. Start Classifying. Are you tired of blindly adding massive blocklists to your Pi-hole and hoping they don’t break your favorite services? Most DNS blocking is a “black box”—you either block everything or nothing.

Enter Donut-Hole.

Donut-Hole is an AI-powered sidecar that gives your Pi-hole a brain. It doesn’t replace your setup; it enhances it. By pulling your DNS logs and classifying every domain using a Local LLM (via Ollama), Donut-Hole lets you audit your network traffic with total clarity.

🚀 Key Features:

• AI Classification: No more “what is this cryptic URL?” Donut-Hole explains what domains actually do.
• The “Sidecar” Approach: Keep your Pi-hole exactly as it is. Donut-Hole works alongside it to generate custom blocklists.
• Vector Memory: Using vector embeddings, the system learns from your previous decisions. The more you use it, the smarter it gets.
• Review UI: A clean interface to audit traffic and export a hosts-format blocklist back to Pi-hole.

🛠️ Project Resources: 👉 GitHub Repository: https://github.com/santhoshjanan/donut-hole/

📚 Prerequisites & Tutorials: If you are starting from zero, check out these excellent setup guides:

Pi-hole Setup (Full Walkthrough): https://www.youtube.com/watch?v=Z_sVjxu9LjE

Ollama Setup (Local AI Engine): https://www.youtube.com/watch?v=N4haIG4kWN8

⏱️ Timestamps: 0:00 The premise 2:17 What is Pi-Hole? 6:45 What is Donut-Hole? Introduction 7:50 How Donut-Hole works? 10:33 Basic architecture 13:28 Demo 18:00 Linking Donut-Hole to Pi-Hole 23:53 How you can set this up? Github Repo 25:15 Wrap-up

#PiHole #SelfHosted #HomeLab #Ollama #LocalLLM #NetworkSecurity #Privacy #DonutHole