Fictional Worldbuilding: Multi-Agent LLM Collaboration with Hierarchical Context Compression and Iterative Review

2026-07-10

AI Summary

This work addresses the challenges of constructing coherent and creative fictional worlds—namely, context explosion, tension between creativity and consistency, and the absence of automated quality assurance—by introducing AutoWorldBuilder, a multi-agent collaborative system. The framework integrates a structured concept network, a DAG-driven hybrid batching scheduler, four-level context compression (reducing token usage by approximately 90%), an iterative review mechanism, and differentiated temperature control. A dedicated auditing agent and a skill-driven, zero-code extensible architecture further enhance narrative self-consistency and generation efficiency. Experimental results demonstrate that the system achieves a 95.0% success rate across 20 diverse world-building tasks, generating 56–103 conflict-free concepts per run within 18–31 minutes, with over 85% of proposals passing expert review.

Abstract

Worldbuilding, the construction of coherent fictional worlds, is a foundational task in game design and literary creation. Large Language Models (LLMs) offer new possibilities for automated content generation, but their application to worldbuilding faces three challenges: context explosion that grows linearly with the building process, the tension between creative diversity and content consistency, and the absence of automated quality assurance. This paper presents AutoWorldBuilder, a multi-agent collaborative system that addresses these challenges through five integrated components: a structured concept network with conflict detection; a DAG-based hybrid batch scheduler that groups tasks by semantic locality; a four-layer context compression mechanism achieving approximately 90% token reduction; an iterative review system with specialized Auditor agents that improves proposal pass rates from 42% to over 85%; and a skill-driven agent architecture supporting 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.

Problem

• worldbuilding
• context explosion
• content consistency
• creative diversity
• quality assurance

Innovation

• hierarchical context compression
• multi-agent LLM collaboration
• iterative review
• semantic-locality scheduling
• conflict-aware concept network