Abstract: Rockbursts are a major dynamic hazard in deep coal mining, characterized by sudden onset and severe destructive potential. Effective prevention and control require a deep integration of specialized domain knowledge and real-time monitoring data. Although large language models perform well on general tasks, their application in high-risk vertical domains such as rockburst prevention faces several challenges, including gaps in domain knowledge, limited ability to incorporate dynamic information, and the high cost of full-parameter fine-tuning. To address these limitations, this study proposes a knowledge-augmented domain-specific modeling framework for rockburst applications, built upon a general large language model. We first construct a high-quality domain corpus through knowledge distillation and apply LoRA-based fine-tuning to improve model adaptability. We then introduce a retrieval-augmented generation (RAG) mechanism to dynamically integrate static regulatory documents—such as industry standards and technical guidelines—with real-time monitoring data. To systematically evaluate model performance, we developed a domain-specific test set encompassing four dimensions: mechanisms and theory, monitoring and early warning, prevention and hazard mitigation, and management and emergency response. Experimental results show that the fine-tuned domain model achieves an overall accuracy of 93.01%, substantially outperforming general models of comparable size. Based on the proposed domain-specific model, three task-oriented intelligent agents were developed: a knowledge question–answering agent, a seismic source analysis agent, and a daily report generation agent for rockburst risk warning. These agents were further integrated through a multi-agent collaborative framework to create a decision-support system for rockburst management, enabling automatic integration, intelligent analysis, and structured reporting of heterogeneous monitoring data such as microseismic and stress measurements. Field deployment at a rockburst-prone mining face in Inner Mongolia demonstrates that the model can integrate regulatory documents with real-time data to generate clear, mechanism-based, and actionable recommendations for issues such as floor heave and high-energy microseismic events. It shows strong advantages in regulatory interpretation, contextual adaptability, and engineering practicality. This study offers a new technical pathway for enhancing the intelligence of rockburst prevention and reducing safety risks in mining operations.