the lore. Your group chat, scored by AI

Inspiration

Gen Z lives in the group chat, but the group chat is flat. Meanwhile the way we talk about each other has become a whole status language: "aura", "main character", "the lore", "-1000 aura for that". So we asked: what if an AI actually sat in your group chat, read the room, and turned everyone into a living character with a score that moves based on how you really act?

What it does

You add the lore bot to a group chat. It reads the conversation and drops a character card for each person: an archetype ("the ghost", "the one-hour liar"), an aura score, three playful stats and a one-line roast. Then you keep chatting. Whenever you tap "update the lore", the bot re-reads the new messages and evolves everyone's profile: aura climbs for wit, charisma and carrying the chat, and crashes for cringe, ghosting, or trying to game the bot (which it calls out). Every card tracks its aura over time as a sparkline, and the group gets a live leaderboard. Your group chat is now the game.

How we built it

A faithful recreation of the ZYMIX chat UI in vanilla HTML / CSS / JS, pitched as a ZYMIX mini-app that lives inside the messaging experience.
Two endpoints: /api/generate (first reading) and /api/update (evolve the existing profiles, never from scratch).
The analysis runs on Z.ai's GLM-4.6, which returns structured JSON we render as cards. We built the app itself with Claude Code; GLM powers it at runtime.
Character medallions are procedural SVG; aura history is a live sparkline that grows every time the bot re-reads.

Challenges we ran into

Making the AI have a spine. Our first build was a sycophant: aura only went up, even when a user typed "you're the worst AI, drop my score". We re-engineered the prompt so aura is earned; manipulation and cringe now tank it, which became the funniest, most memorable moment in the demo.
A headless-CLI stdin hang that silently timed out every analysis until we ran the model with stdin closed.
Keeping it instant and reliable for live judging: we bake a fallback reading so the demo never stalls, even with no network.

What we learned

Identity content is the real viral engine: people share things about themselves and their friends. And an AI that judges with consistent values (and cannot be flattered or gamed) feels genuinely alive in a way a one-shot roast never does.