Safelora: A Real-Time Multimodal Multi-Agent Advisory System for Context-Aware and Safety-Compliant Plant Diseases Management

Plant disease management requires timely, accurate, and safe decision-making, yet existing advisory systems often rely on static knowledge and single-agent reasoning, limiting their ability to adapt to real-time field conditions and prevent unsafe recommendations. To address this gap, Safelora presents a real-time multimodal multi-agent advisory system for context-aware and regulation-compliant plant disease management. The system integrates a CNN-based Image Agent for visual disease diagnosis, a Context Agent for iterative knowledge retrieval, a Safety Agent for regulatory validation, and a fine-tuned Large Language Model as the Generative Agent, while incorporating real-time factors such as local weather conditions and crop growth stage information. Experimental evaluation demonstrates strong disease recognition performance and high semantic quality of generated recommendations based on BERTScore, with system-level assessment using RAGAS showing high retrieval accuracy and factual consistency. These results indicate that the proposed system can support reliable, safe, and field-aware decision-making in real agricultural environments. By combining multimodal perception, iterative retrieval, and deterministic safety validation within a unified multi-agent framework, Safelora offers a distinct and practical approach to plant disease advisory generation beyond existing static and single-agent solutions.