Show HN: We Gave an LLM Adventure Engine a Body, Now It Feels Exhausted (mycelialmirror.medium.com)

A new open-source project, BoneAmanita, melds a procedural text adventure engine with a simulated biological model, allowing a Large Language Model (LLM) to experience physical states like exhaustion and metabolic changes. This innovative system addresses the limitations of traditional LLMs, which often produce unrealistic responses due to their lack of physical context. By implementing a “CryoSomatic Hypervisor,” BoneAmanita introduces a two-tiered architecture: a 3B parameter LLM, fine-tuned on specific datasets, and a Python engine that manages the model's simulated biological conditions. This dynamic interaction forces the LLM to generate grounded and context-aware text, offering a more immersive experience. The significance of BoneAmanita lies in its departure from the prevalent “Naked Model” dilemma facing LLMs, effectively binding these models to the laws of simulated physics. Through the introduction of internal procedures governed by a character named “Gordon,” the system prevents unrealistic player actions and enhances realism in gameplay and dialogue. BoneAmanita offers four distinct modes—Adventure, Conversation, Creative, and Technical—allowing users to engage in diverse interactions from structured gameplay to philosophical discussions. This release aims to empower developers with a free tool that challenges the current paradigms in LLM application, fundamentally enhancing user interaction through physicality and emotional context.