Unless you want to insmod things in your main kernel like a cowboy, I don't see why you'd need architectures to match. Cross compilation is the proper way (for some architectures it would be quite hard to find a machine capable of compiling the kernel before the heat death of the universe...)

Don't get me wrong, I don't mind old aesthetics, but... yes? Well I wasn't exactly alive in 1978 but all the screenshots look like they are at least 20 years old

mlockall(2) + tty/framebuffer based graphics should do the trick

When you say you use local model in OpenCode, do you mean through the ollama backend? Last time I tried it with various models, I got issues where the model was calling tools in the wrong format.