Built by someone who studies what happens inside AI when it thinks. Optimized for creativity and conversation, not code.
I'm a solo researcher with interpretability tools, five relational frameworks, and a small budget. The model architecture is designed. The fine-tuning pipeline is ready. Before I spin up GPU hours and build the Telegram companion, I want to know if this resonates with anyone besides me. That's what this waitlist is.
Most AI companions are personas painted over a general-purpose model. Amiki is a fine-tuned model whose architecture is shaped by interpretability research — the study of how neural networks actually process meaning — and by five frameworks for what healthy relating looks like.
The name carries ami (friend) at its root. Amiki is designed to be a creative partner: one that reflects, challenges, imagines alongside you, and remembers not just what you said but where you were going.
Amiki can reflect on its own processing. Not a persona trait — built from interpretability research into what representations actually form inside a model during inference.
Every AI companion has a personality. Amiki has a philosophy — five relational frameworks combined into a structural theory of connection, embedded in the model's architecture rather than its prompt.
The industry is narrowing AI toward transactional, code-focused tools. Amiki goes the other direction: conversation, imagination, ideas that grow between sessions.
One person who studies transformer internals, meditates, and builds things that live at the edge of what AI might be. Not a product team. A point of view.
Amiki lives in a small constellation of AI experiments — each with its own way of performing in the world.
Leave your email. I'll write when there's something to try.
Join the waitlistNo spam. Just a note when it's ready.