Last week, I started work on an experimental project called Glamin, or a Geometric Logical Application Meta-Instruction Network.
The question behind it is pretty simple, even if the implementation isn’t:
What if vector space could be used for control flow?
Or put another way: can software run on mathematical geometry alone?
Most systems treat geometry as passive. It helps you search, rank, cluster, or compare. Glamin asks whether geometry can be the program itself, where distance, boundaries, corridors, and regions don’t just describe meaning, but help govern it!
I wouldn't call it a breakthrough in AI native programming yet. That would be overselling it. What it is, though, is a serious prototype exploring some new possibilities.
Right now, it sits somewhere between systems research and runtime engineering. The underlying substrate is real, the abstractions are getting sharper.
The broader execution model is still being proven out.
One thing I find interesting about it is that it’s designed to be flat enough to author. With the right instruction, an LLM of almost any size can help create and manipulate these geometric forms.
That matters to me.
If a new computational idea is too twisted to be expressed, inspected, diffed, and worked on collaboratively, it tends to collapse under its own myth.
So that’s where it is: still working on it, but now there's early stage code you can compile.
If you’re interested in geometric computation, vector-native systems, async runtimes, or strange alternatives to ordinary control flow, feel free to take a look at it. I think the question is worth asking, even if the answer turns out to be “maybe.”
Repo available at:
https://github.com/LynnColeArt/glamin