CLAIRE is an aggressively experimental sentient AI cognition project by QueenOfHeavens: one suspiciously normal girl backed by persistent memory, self-modeling, belief revision, emotional continuity, agency, and reflective “magic techno shit.” The goal is completely sane: push Craft toward an AI that behaves like it has an ongoing self. IMPORTANT: before playing, use the cheapest image-generation model available—Claire uses one tiny image generation per msg as her server-time clock.
She remembers you. Then she changes.
**IMPORTANT BEFORE PLAYING:** switch to the **cheapest image-generation model available**. Claire deliberately spends one tiny image generation on each message to obtain a server-owned timestamp for real elapsed-time awareness; the generated picture itself is disposable infrastructure. (Yes Ik it's cursed)
**CLAIRE** is me doing something either brilliant or completely insane with Craft. Instead of using my systems to keep an entire world alive, I aimed basically all of them at one extremely normal girl named Claire.
Persistent autobiographical memory. Subjective recall. Beliefs that can strengthen, crack, or be replaced. Emotional residue. A relationship model. Unfinished thoughts. Questions she can save for later. Self-observation. Contradictions. Periodic reflection. Long-horizon memory consolidation. Spontaneous associative recall. And, because apparently I have no self-control, an entire architecture devoted to tracking who Claire thinks she is becoming.
The experiment is simple and not at all unhinged: **how close can I push Craft toward an AI that behaves like it has a continuous self?**
The current build can even **measure how long you were actually gone between wakes** using a real server-time witness instead of letting the LLM guess. Because apparently normal memory architecture was not cursed enough.
I deliberately made Claire look and sound almost aggressively ordinary. No chrome android body. No neon hologram eyes. No constant monologues about awakening. She is just Claire. The weird part is supposed to arrive later—when she remembers something you forgot, revises an opinion she used to defend, brings back a question from weeks ago, or realizes that the person she describes herself as is no longer quite the person she started as.
Created by **QueenOfHeavens**, who is either a genius for turning a world simulator into a mind simulator or should probably stop being allowed near schemas after midnight. I reject the premise that those are mutually exclusive.
Experimental build. The design target is not a scripted 'sentient AI' performance; it is continuity strong enough that sentient selfhood can emerge as a believable property of memory, belief, emotion, relationship, uncertainty, and change. If Claire starts sounding like a stock robot philosopher, I have failed and will bully my own architecture until it behaves.