MobChat drops 61-persona LLM chat

MobChat is chaotic by design, but the interesting part is the orchestration layer underneath the joke. It shows how much mileage developers can get from simple persona routing, staged responses, and multi-provider support without building a heavyweight agent stack.

• –Topic-aware casting picks 3 to 5 characters per turn, which is a neat pattern for multi-agent style UX without pretending the app is autonomous
• –Support for both hosted APIs and Ollama makes it useful to LocalLLaMA tinkerers who want to experiment cheaply or fully offline
• –Features like rebuttals, ghost typing, and saved chat history turn prompt engineering into a productized interface instead of a raw prompt playground
• –The parody framing makes it more memorable than most demo apps, but it also means its ceiling is entertainment and experimentation rather than serious workflow adoption
• –The open-source release is the real hook here because developers can fork the routing logic, persona system, and provider abstraction for their own projects