Questions here go to a small language model running on an old desktop in my house. There's no data center or paid AI service behind it, just that one computer, so it's slower than you're used to and it only answers while the computer is on. Run a starter cell or type your own below.
Answer library—answers it writes ahead of time while nobody's asking
The trip your question takes
1Your browser → this siteVercel hosts the site and holds on to your question. It never runs the model.
2A queue in the middleThe desktop checks the queue and pulls questions down. Nothing on the internet can connect to it directly.
3Safety railsPrompt-injection attempts, impersonation and anything that would commit me to work get a fixed reply, and the model never sees them.
4The model reads a profileIt gets one document about me. If something isn't in there, it's supposed to say it doesn't know.
5Back to youUsually 10 to 20 seconds, or right away if someone asked the same thing before.
The wait is real. It's a small model on four CPU cores in my house, with no rented GPU anywhere in the chain. When nobody's asking it works ahead, writing answers to the questions people tend to ask, so those come back as soon as you hit enter. Ask it something new and you're waiting on that old desktop to actually think it through.