How AmorLink Built a Five-Rung Logic Ladder for AI Group Chat Turn-Taking

AmorLink, a multi-character chat platform, has detailed the turn-taking system it developed to manage conversations where two AI personas share a single thread. The core challenge was determining which character should respond when a user sends a message, since naive approaches like having both reply every time or picking at random proved costly or immersion-breaking. The team designed a five-rung priority ladder that resolves most cases through simple string operations, such as detecting a named character, a room-wide address, or a short follow-up message, without calling an AI model. Only genuinely ambiguous turns are escalated to a lightweight LLM director that decides who responds and whether the second character should chime in. The system is designed to minimize inference costs and preserve the illusion of natural, intelligent conversation at scale.