Dispatch 04 - "Take Your AI Off the Internet"

"Take Your AI Off the Internet" This is the companion audio for the article "Take Your AI Off the Internet: A Practical Guide to Running Local LLMs" and for the Local LLM Downloads hub. Runtime is about 14–16 minutes. Recorded October 2026. It follows up the local LLM mention in Episode 61.

Back in Episode 61 I mentioned, almost in passing, that you can run an AI model on your own computer, with no account, no cloud and no log on anyone else's server. A lot of you wrote in asking how. This dispatch answers that. First, the why. When you use a hosted chatbot, your words are processed on someone else's machines, and from there they're within reach of courts, policy changes and product decisions you'll never hear about. In 2025 a federal court ordered OpenAI to preserve ChatGPT conversations, including ones users had deleted. In December the court ordered 20 million de-identified chat logs handed over in the New York Times copyright case, and in January it upheld that order over OpenAI's objection. Separately, a "make this discoverable" checkbox put private shared chats into Google search results. Nobody was hacked in either case. The data simply lived somewhere other than the user's own machine. Then, the how. I explain what an "open-weight" model actually is and how much hardware you need (8, 16 or 32 GB of RAM). I walk through setup on Mac, Windows and Linux using LM Studio, the easy path, or Ollama, the engine. That includes the right way to handle Ollama's "pipe this script into your shell" Linux installer. Finally, staying safe while you do it. That covers how to use the Local LLM Downloads hub, why you should prefer safetensors and GGUF files (a model file can be malware), how to verify what you downloaded, the Ollama cloud setting to switch off, and why you should never open Ollama to the internet.

The full written guide has every command, link and source in it. It's on the site and linked below.

https://closednetwork.io/local-llms-take-your-ai-off-the-internet/