AutoWorldBuilder uses hierarchical context compression and iterative review for coherent worldbuilding

The paper introduces AutoWorldBuilder, a multi-agent collaborative system designed to address context explosion, consistency, and quality assurance challenges in automated fictional worldbuilding.

• A four-layer context compression mechanism achieves approximately 90% token reduction.
• An iterative review system with specialized Auditor agents improves proposal pass rates from 42% to over 85%.
• Experiments using GPT-OSS 120B and DeepSeek v3.2 demonstrate a 95.0% success rate across 20 tasks.
• The system generates 56-103 self-consistent concepts per world in 18-31 minutes with zero-conflict delivery.

The authors argue that these architectural patterns, including layer-as-budget compression and semantic-locality scheduling, transfer to broader knowledge-intensive multi-agent LLM applications.