The character has four traits — warmth, edge, wonder, guard — that shift as you talk. Every shift has a cause: something you said, or something it said back. Big moments fade slowly; small ones fade fast.
The engine runs inside the app. The talking is done by an AI service you connect once in settings — your choice of three, listed below.

Warmth, edge, wonder and guard, each on a 0–20 scale, moved by what's said — never by random chance.
Trait bars, need gauges and active dimensions update on screen with every message. You can watch the mood move.
Each change fades at its own speed: a routine remark wears off in a few turns, a heavy moment keeps influencing the voice much longer.
Named tangles like jealousy with a resting level — they spike when triggered and settle back to a simmer, not to zero.
Floors and ceilings cap how far a trait can be pushed. Soft guards lose strength under sustained pressure and recover on calm turns.
Only the three most active dimensions lead a given reply; the rest continue quietly in the background.
The character's replies are scored too, at half weight — what it says affects its own state.
Connection, purpose, safety and clarity drain and refill with events, and quietly bias the traits.
In the app: Settings → Connect with Pollinations. You sign in on Pollinations' own site with your existing account, approve, and you're back in seconds. Or paste a key.
Install Ollama on your computer and enter its address in settings. The app and your PC must be on the same network.
If you use Claude Code: download the bridge script, run it on your PC with Python (it uses your existing Claude sign-in), and enter your PC's address in settings.