BatLLM

BatLLM combines a playable local desktop game with a read-only game analyser, model-management tools and a separate research recorder and verifier. Players write prompts that control game bots through local large language models.

The project is intended for practical and critical exploration of prompt quality, model behaviour, grounding and replay. Its ordinary graphical session format remains distinct from the explicit research trace used for reproducible experiments. A paper on verifiable grounding and operative replay in the system was accepted for IEEE URUCON 2026.