KYOTO MOON

Putting retrieval in the hands of AI users

AI is very good at generating information.

But sometimes we don't need a new answer.

We need the information we already have.

THE PROBLEM

AI systems can lose context, reconstruct work that already exists, drift from important instructions, and use computing resources to generate information that could simply be retrieved.

That can mean lost work, inconsistent answers, less user control, and unnecessary cost.

It can also mean unnecessary use of electricity and the water needed to cool data centers.

THE IDEA: RETRIEVAL OVER GENERATION

AI already uses retrieval through technologies such as RAG — Retrieval Augmented Generation — to find and use stored information.

Kyoto Moon asks a simple question:

Why not take that idea further and put it in the hands of the users?

When reliable information already exists, retrieve it rather than regenerate it.

WHAT WE'RE BUILDING

Retrieval Tools

Simple tools that help people save, recover, and reuse their own AI work instead of asking AI to reconstruct it.

Steady Cards

Small information objects stored outside the AI that can preserve things that should not casually change:

• important instructions

• decisions and constraints

• source information

• definitions and assumptions

• user-defined limits on what an AI is authorized to do

The AI can remain powerful and flexible.

What needs to stay steady can stay outside it.

WHY IT MATTERS

Retrieval Over Generation could contribute to:

Better continuity — Keep work from disappearing when AI loses context.

Greater user control — Keep important information and instructions under the user's control.

More reliable AI — Retrieve authoritative information rather than repeatedly reconstructing it.

Provider independence — Make important information less dependent on one AI company or model.

Safer AI delegation — Preserve boundaries between what a user wants and what an AI is actually authorized to do.

Lower resource use — Reduce unnecessary computation, with the potential to lower cost, electricity use, and cooling-water demand.

WHAT COMES NEXT

Build → Test → Measure → Share

Kyoto Moon has moved from the original idea into designs, documentation, safety principles, and early technical work.

Now we need working prototypes.

We plan to build and test small demonstrations of Retrieval Over Generation and Steady Cards, measure what they actually do, and publish what we learn.

The goal is not simply to say:

“We think this works.”

It is to be able to say:

“Here it is. Try it.”

THE LONGER VIEW

Kyoto Moon's goal is to build a sustainable business around affordable, user-controlled AI tools.

But the opportunity may be larger than any one company.

If these ideas work, we want researchers, developers, small businesses, nonprofits, and large technology companies to be able to test them, challenge them, improve them, and find uses we haven't imagined.

Some of AI's important problems may not require more AI to solve them.

Sometimes the answer may be simpler technology, used in a different way.

ABOUT THE FOUNDER

Marcia Coulter founded Kyoto Moon LLC after decades of work in and around technology, including technical writing, instructional design, project management, and analysis.

Kyoto Moon began as a toy company. In 2025, an AI system repeatedly lost work from a large project. A simple attempt to solve that problem — save the work outside the AI and retrieve it when needed — changed the direction of the company.

Kyoto Moon now explores what else becomes possible when we choose retrieval over regeneration.

Kyoto Moon LLC
Open Knowledge with Roots
KyotoMoon.com