🤓 Latest Submissions
agentSpam
agentSpam is a multi-agent system that does the opposite of what most LLM apps do. Instead of asking one big model for one polished answer, it spawns up to 90 specialist agents in a recursive tree, lets them disagree, and then synthesizes the result back up. A single Brain agent decomposes your question into subtasks and dispatches them to specialists it invents on the spot — a Risk Hawk, an Ethical Strangler, a Cost Auditor, each one naming itself before working. Those specialists decompose further until interns at the bottom of the tree answer in 1-3 sentences. Every layer synthesizes its children on the way back up. In demo mode, the Brain stages a real debate: The Bull argues for, The Bear argues against, three rounds of rebuttals each, and The Judge renders a single-sentence verdict that the browser speaks aloud. Agents can also propose one-click Google Calendar invites and email drafts that fire in your already-logged-in browser — no OAuth. Every run runs side-by-side with a single Claude call on the same prompt, so you can see the difference between one opinion and ninety. Export the whole debate as an HTML report styled like a photocopied punk zine.
19 May 2026
