The fast pace of Artificial Intelligence (AI) development brings disruptive possibilities in managing personal finance, but current systems are confined by the inflexible architecture and the limitations of not having adaptive reasoning, contextual awareness, or intelligible decision-making. This paper presents the Finance Utility LLM, a novel architecture that utilizes a multi-agent architecture enabled by Large Language Models (LLMs) to build an intelligent, conversational financial assistant. The proposed architecture overcomes the fundamental limitations of current rule-based applications by coordinating domain-specific agents for automated expense tracking, Optical Character Recognition (OCR) for receipt digitization, dynamic financial knowledge retrieval, and personalized financial planning through a single natural language interface. By interfacing with digital payment ecosystems, and utilizing structured prompt engineering, the architecture translates complex user intents into coherent financial actions while supporting robust data privacy using a hybrid deployment model. The architecture is also designed to improve accuracy in decision-making through a multi-stage validation approach, improve operational reliability through containment of modular faults, and increase financial literacy by embedding Explainable AI (XAI) features that provide understandable justification for any recommendations provided. This work provides a holistic architecture that bridges the gap between general-purpose capabilities of LLMs and the nuanced requirements of financial applications and serves as a foundation for further development.