Integrating Large Language Models (LLMs) with climate model data, scientific literature, and unstructured text enables a new generation of climate information systems that deliver accurate, localized, and context-aware insights. Our primary objective is to develop and evaluate ClimSight, a scalable platform that turns complex heterogeneous data into actionable information. We augment LLMs with Retrieval Augmented Generation, a method that retrieves relevant climate models and reports at query time to ground responses. An agent-based architecture orchestrates specialized modules that route and process user queries with task-specific tools. Real-world evaluations compare multiple LLM configurations and analyze trade-offs between speed, cost, and accuracy. Results show improved scalability and precision in climate assessments, democratizing access to localized information. This paradigm shift equips stakeholders in agriculture, urban planning, disaster management, and policy with effective tools for forward planning and risk management.