AGORA-AI is an evidence-grounded recommendation and what-if assistant for municipalities and city-service operators, delivered as a containerised AI layer on DOTSOFT’s NOON-IoT smart-city platform. A user describes an urban challenge in natural language, such as traffic congestion or excessive energy consumption. AGORA-AI searches a knowledge graph linking urban problems, interventions, city contexts and measured outcomes from peer-reviewed evidence and operational city data. Its hybrid similarity engine identifies genuinely comparable cases, while the scenario engine estimates likely impact ranges and explicitly communicates uncertainty. A European LLM, constrained through retrieval-augmented generation, produces ranked recommendations with traceable sources. The system does not make autonomous decisions: the administrator retains full authority. Built on NGSI-LD, FIWARE and open APIs, AGORA-AI is portable, interoperable and suitable for accountable public-sector use.