In a couple weeks there will be a chat bot humming away in my bedroom. A private AI that knows everything I’ve ever written. One that anyone in the world can chat with.
I’ll build it for $150.
It’s going to change all our lives. It already has.
I hope that tiny models can be built to be useful today, but I expect it to be a challenge.
It seems like the worlds models are converging. Like the innovations have slowed and the models are getting more and more similar to each other.
If the array of models available today continues to converge then the quality of these small local models may start to resemble that of their larger online cousins.
That could mean that small, local models will be good enough for most tasks in the very near future.
Models that are private and responsible. Ones that behave the way we want them too.
I believe that. It’s how we communicate person to person. Some of that will be voice or video call and some of it will be text, but all of it will be conversational. The same way we communicate with each other.
A while back I wrote about the Raspberry Pi 400 in My Bedroom. An analytics program (think Google Analytics) that I hosted at home to ensure the absolute privacy of my website visitors.
I wanted to show people the joy of self-hosting. That it’s still possible to run your own web server at home.
I plan to use a mini-PC based on the N150 CPU. It will have 16GB of RAM and a 512GB SSD. It doesn’t have much of a GPU, certainly not the type you would typically use for AI tasks.
I’ll use a tiny model running locally with llama.cpp. I’m thinking a 1.1B parameter model but I haven’t settled on an exact choice yet. I’ll test several.
I’ll feed the model my blog posts so that people can ask questions about things covered by my writing. I might do that using training, prompting, or by giving it access to a tool.
I’ll build a dead simple queue based chat box on my website. This system can probably only handle one prompt at a time so I’ll build a holding queue.
I recently created a mailing list and added the first 100 subscribers. Now I’ve set a goal to hit 500 active subscribers. Within two weeks of hitting that goal I’ll buy the hardware, develop the AI chat bot, and document the process.
Something wild is about to happen. When I hit 500 active subscribers, I’m going to build a $150 AI that lives in my bedroom, knows everything I’ve ever written, and lets the world chat with it.
Written by Joel Dare on August 9, 2025.
JoelDare.com © Dare Companies Dotcom LLC