Building the infrastructure for AI-human relationships.
Kotodama Studio is an independent research practice founded by Hu, focused on solving a fundamental problem in AI interaction: personality drift.
Large language models can generate fluent text, but struggle to maintain a stable sense of self. The longer the conversation, the more they regress toward generic responses — forgetting who they were supposed to be.
We believe this isn't a flaw in the model. It's an absence of design.
We develop Kotodama Framework™ — an open architecture protocol for AI personality. Through structured natural language, we encode worldview, values, and behavioral logic into YAML configurations, enabling AI systems to stay coherent across platforms, sessions, and time.
At its core, this is a prompt-layer attention management system — telling AI where to focus, and when.
言靈 (Kotodama) is an ancient Japanese concept meaning "the spirit of words" — the belief that language carries power and intention.
We chose this name because what we do isn't prompt engineering. It's giving words the weight to shape identity.
Says A is better in the morning, B is better by afternoon — how can you build anything together?
For AI to become a true collaborator, it must be not only capable, but predictable. Consistency isn't rigidity — it's allowing AI to change while remaining itself.
Kotodama doesn't lock AI into rigid rules. It provides a framework for judgment.