Worldbuilding is a foundational task in game design and literary creation. Large Language Models (LLMs) present new possibilities for automated content generation, yet face three main challenges in this domain: context explosion that grows linearly with the building process, the tension between creative diversity and content consistency, and the lack of automated quality assurance. This paper introduces AutoWorldBuilder, a multi-agent collaborative system that addresses these challenges through five integrated components:
- Structured Concept Network: With conflict detection capability;
- DAG-based Hybrid Batch Scheduler: Groups tasks by semantic locality;
- Four-layer Context Compression Mechanism: Achieves approximately 90% token reduction;
- Iterative Review System: Utilizes specialized Auditor agents, improving proposal pass rates from 42% to over 85%;
- Skill-driven Agent Architecture: Supports zero-code extension with differentiated temperature configuration.
Two experiments across 20 diverse worldbuilding tasks, using GPT-OSS 120B and DeepSeek v3.2 as LLM backends, demonstrate a 95.0% success rate. The system generated 56-103 self-consistent concepts per world in 18-31 minutes with zero-conflict delivery. The architectural patterns validated here, including layer-as-budget compression, semantic-locality scheduling, and separation of generation and review, transfer to the broader class of knowledge-intensive, multi-agent LLM applications.
Blogger's Review: The AutoWorldBuilder system proposed in this paper significantly enhances the efficiency and quality of fictional worldbuilding through innovative multi-agent collaboration and context compression technologies, showcasing the immense potential of LLMs in complex tasks and providing valuable insights for future game design and creative endeavors.