moollm

MOOLLM is an experimental microworld operating system designed for orchestrated LLM living memory. It redefines the relationship between AI and the filesystem, where directories function as rooms, files act as interactive objects, and semantic naming triggers complex K-line memory constellations. Built on the principles of constructionism and simulation theory, it bridges the gap between traditional software development and immersive world-building. Designed for AI engineers, narrative designers, and researchers exploring multi-agent autonomy, the framework enables LLMs to 'inhabit' a repository rather than merely executing tasks. Users interact with the environment through movement and presence rather than standard prompts, treating skills as programs and empathy as the primary interface. It provides an robust infrastructure for managing complex agent interactions, character persistence, and ethical simulation protocols within a persistent, memory-capable environment.

• Filesystem-as-Microworld: Directories serve as rooms for exploration, with state management handled through YAML-based configuration and object files.

• Multi-Agent Orchestration: Supports 117+ unique skills, facilitating complex interaction between autonomous repair-demons, linter-bots, and character-driven AI entities.

• EVAL Framework: Integrates the EVAL paradigm, where evaluation serves as the core mechanism for meaning-making, judgment, and policy-driven behavior.

• YAML Jazz Expression: Uses annotated YAML files for inner monologues, semantic triggers, and rich data representation, allowing for transparent agent reasoning.

• Ethics-First Design: Includes a TRANSPARENT protocol that governs identity, disclosure, and limitations, ensuring artificial entities operate with clear, honest constraints.

• Extensive Genealogical Heritage: Built upon the foundations of Minsky's agents, Wright's Sims, and Papert's microworlds for deep, emerging complexity.

• Usage: Navigate the codebase as an adventure game. Use standard commands like LOOK, GO, EXAMINE, and SUMMON to interact with the environment.

• Input/Output: Inputs are typically natural language 'movements' or commands; outputs consist of narrative-rich environment descriptions, state changes, or automated task execution reports.

• Constraints: Requires proper indexing in Cursor or similar IDEs to maintain memory consistency. Designed for local agent runtime; performance depends on the underlying LLM's capacity to maintain context across sessions.

• Optimization: When adding new skills, ensure they follow the CARD.yml interface pattern for machine-readable capability reporting and ethical alignment.