Building My Own Private AI, One Agent at a Time
Running local LLMs at home with LM Studio, Llama, and Gemma, and building a cybersecurity agent workflow, all for privacy and control of my own data.
I’ve gone down a rabbit hole I don’t want to climb out of: running large language models locally, on my own hardware, fully under my control.
Where it started
It started with LM Studio and Llama. Pulling down a model and running it on my home PC, no cloud, no account, no data leaving the house, hooked me immediately. Then I got into agents, including the ones PewDiePie showed off in his local-AI setup, Hermes and Odysseus, which pushed me past just chatting with a model.
My own private AI
Now I host an open-source Gemma model on my home PC as my own private AI, and I’m building out an agent workflow aimed at cybersecurity, something to help me practice running detections and work through investigations. That project is where I learned what an MCP server actually is and how agents talk to tools and data behind the scenes.
Why local?
The real reason I keep at it is privacy. I care about where my data goes, and running everything locally means I stay in control of it instead of handing it to someone else. This space moves fast, and I’m always chasing the next bit of AI news to keep up.