Recently, large language models (LLMs) have demonstrated a high level of intelligence, with extraordinary common sense, reasoning, and planning skills that frequently provide insightful guidance. The prospects are vast for the application of these capabilities in the operation, maintenance, and management of low-altitude unmanned aerial vehicles (UAVs). However, general LLMs lack specific expertise in UAVs, making it challenging for them to provide expert and insightful guidance or recommendations. To bridge this gap, we present the UAV dual-agent—a framework using large language models and knowledge graphs to address UAVs operation and maintenance issues, thereby providing in-depth decision support for urban low-altitude UAV through natural language dialogue. By constructing a UAV doctree, utilizing a knowledge graph to correct and supplement the doctree, and employing collaborative reasoning with LLMs, we have achieved a 40% increase in the accuracy of operational decisions compared to general LLMs.