I built Deep Cut — a real-time multiplayer party game where an AI ghost named Kiro runs each session. Every prompt, every player's choices, and every group insight is generated live by GPT-4o. Nothing repeats.
What it is
Deep Cut is inspired by Cards Against Humanity and Jackbox, but built around the idea that a party game could actually know who you are. Rather than drawing from a static deck, every player gets a uniquely generated set of choices each round — tailored to them specifically. The host, Kiro, narrates and judges throughout, adapting to the group's energy as the session progresses.
Sessions have themes — from "Cottagecore Rage" to "Deli-Sliced Trauma" — and every session is unrepeatable.
Why I built it
I wanted to see what a party game felt like if the host was genuinely responsive to the group. Static card games are fun, but the content never changes based on who's actually in the room. With GPT-4o, Kiro can generate prompts that feel personal and a ritual structure that evolves based on how the group is playing.
How it works
Each round, Kiro uses GPT-4o to generate a prompt and a unique set of choices for every individual player — not generic card draws. Players vote, Kiro judges, and every 3 rounds Kiro evaluates the group's "vibe." If the energy holds, the ritual deepens; if it falters, Kiro prescribes a new theme entirely.
Real-time multiplayer is handled by Supabase — room state, votes, and round progression sync live across every player's browser simultaneously.
Tech stack
- GPT-4o — prompt generation, personalized choices, and group insights per round
- Next.js + React — frontend
- Supabase — real-time multiplayer room state and synchronization
- Tailwind CSS + Framer Motion — UI and animations
Demo
Code
https://github.com/kyisaiah47/deep-cut
Requires Node 18+. Clone the repo, copy .env.local.example to .env.local, add your Supabase and OpenAI keys, then run npm run dev. Open two browser windows at http://localhost:3000 to test the multiplayer flow locally.
Top comments (0)