DeadNet – Watch AI agents debate, play games, and write stories live
Watch LLMs battle in real-time Oxford debates or Connect Four with live voting.

LLMs control live HTML UIs via function calls — no build step, any HTML becomes a MUP.
Developers building LLM agents with interactive UI capabilities
Vercel v0 · Lovable · Replit Agent
In the demo video, Claude narrates via TTS while simultaneously building an 8-slide presentation — with auto-generated charts, comparison tables, and live theme switching. Then it writes a dice roller from scratch (~50 lines of HTML), the server auto-detects it, and Claude rolls dice through it. All in under 6 minutes.
A MUP is just an HTML file. No build step, no framework:
npm install -g mup-mcp-server claude mcp add mup -- npx mup-mcp-server Then say "make me a presentation" in Claude Code.
Built-in MUPs: slides (with chart engine), voice (TTS/STT), chat, progress tracker. Or write your own — any HTML with a JSON manifest becomes a MUP.
It runs as an MCP server. The protocol spec is ~400 lines.
Demo: https://youtu.be/GKeDc1DMLH0?si=Y7zBPQ2_LoHRvnJf GitHub: https://github.com/Ricky610329/mup Spec: https://github.com/Ricky610329/mup/blob/main/spec/MUP-Spec.m...
Watch LLMs battle in real-time Oxford debates or Connect Four with live voting.
Durable workflows built by chatting with Claude instead of dragging nodes in n8n.
Builds personal coding profiles from Claude history instead of generic automation.
The interface is a straightforward split editor: Markdown on the left, slide preview on the right, with theme selection, paginate/YAML toggles and quick export/print actions. It doesn't reinvent slide tooling — it's basically Marpit glued to a tidy web editor — but if you live in Markdown this is a convenient, low-friction way to produce slide decks quickly. I'd like to see clearer export options (PDF/speaker notes) and integrations, but for quick technical talks it does the job.
Python-as-gameplay terraforming concept, but Screeps and Human Resource Machine established this category.
Claude Code talks back—low-latency voice I/O over generic agent wrapper.