I was building websites with AI, and every day I found myself explaining something we had already settled. A layout failed in the same way. An instruction disappeared between sessions. A decision that had cost us an afternoon became something I had to defend again the next morning.

At first, I wrote things down. That helped, but only a little. The notes existed; I still had to remember where they were, bring them into the conversation and insist that they mattered. I had become the memory of the system. Working with an assistant meant supervising its ability to remember our work.

The first Lore grew while I was developing Blockchain Lab UAI. I was trying to give decisions somewhere to live. The name came from games: a world has a history, and what happens in it should have consequences. Our work with AI seemed to begin again whenever a session ended.

Keeping a record and applying it are different problems. A document can explain a mistake perfectly while the next session repeats it. That difference became the question behind LUS: when accumulated judgment is available, does it actually change a later decision? Much of the evidence comes from my own projects. Wider validation is still work to do.

The question also had a cultural life. Groundhog Day gave us a picture of repetition; Tales of Berseria shaped Lore Plugin’s visual sensibility. These were things that helped us imagine the work, kept separate from the evidence needed to judge it.

Lore Plugin became the practical arm of that inquiry. A contract sets the work, Lore keeps the judgment behind decisions, and skills help us load, distill and revise it. The kit grew through projects and corrections, including failed websites. Its criteria can change; keeping everything forever would create another problem. Vespi carries the operational half: what an agent may do, what happened and what can continue afterward.

The kernel makes authority, effects and receipts concrete. Tools and storage still depend on the environment. For Find Your Way, the kernel is what we are asking judges to evaluate.

Then we built eleven projects in a week. They are the cases you can open from this desktop: different problems, agreements and limits, including prototypes with fictional data. Their public repositories are the entry point. Documentation and executable code must be judged separately, according to what each repository actually contains.