LIVE

I Built AI Characters That Live Their Own Lives (No API Costs)

Episode 02 · Little Computer People · Interactive Demo

🏠 Live Demo

Watch autonomous AI characters navigate their world — eating, cleaning, playing music, socializing. Powered by hybrid utility-AI + optional LLM reflection. All running locally on dual RTX 5090s.

📊 By the Numbers

API Costs

27B

LLM Parameters

Rooms

126

Furniture Pieces

∞

Messages

Subscriptions

⚙️ How It Works

🧠 Utility AI

Need-based decision engine. Characters track energy, hunger, hygiene, social, fun, and creativity. Each furniture piece has associated actions with priority scores.

💭 LLM Reflection

Optional layer using local Qwen 3.6 27B via vLLM. Adds personality-driven thought, complex decision-making, and natural language narration. Disabled by default for demo.

🏗️ Local Stack

Express server + React client. Runs entirely on dual RTX 5090s. No external API calls. Deployed to Railway for public access.

🎮 What Characters Do

Eat & Drink — Use the fridge, stove, and counter to satisfy hunger
Clean Up — Bathe in the bathroom, maintain hygiene
Play Music — Piano in the living room and studio
Socialize — Interact with other characters in shared spaces
Relax — Watch TV, read on the couch, use the computer
Explore — Move between 6 rooms across 3 floors

🎥 Watch the Video

📹

Episode 02 Video Coming Soon

Full walkthrough with commentary: architecture deep dive, live demo, and what local AI can do right now.

Try the live interactive demo above while you wait.

▶ Subscribe on YouTube

🔧 Technical Details

Component	Details
Language Model	Qwen 3.6 27B (via vLLM, int4 compressed)
Hardware	Dual NVIDIA RTX 5090s
Inference Server	vLLM 0.19.1 with flash-infer + fp8 KV cache
Backend	Express.js with TypeScript
Frontend	React + Vite
AI Architecture	Hybrid: Utility AI (default) + LLM Reflection (opt-in)
Deployment	Railway (server) + Cloudflare Pages (companion site)
Cost per Decision	$0.00 (all local)

▶ YouTube Channel