Accessing legal information in Vietnam is fraught with challenges, primarily due to the sheer volume of legal documents and the complexity of legal language. While Large Language Models (LLMs) present a potential solution, they are often hindered by issues of ‘hallucination’ and outdated knowledge. To overcome these obstacles, this paper introduces LAURA (A Legal Agent Using Reasoning and Acting), a novel agent-based question-answering system tailored for Vietnamese law. LAURA operates on an agentic workflow, leveraging an LLM as its core reasoning engine. The system is built on the ReAct paradigm, which synergizes reasoning with the use of retrieval tools to process legal queries effectively. To enhance accuracy and minimize errors, a custom retrieval module was developed, combining Reciprocal Rank Fusion (RRF) with the BM25 sparse retriever and the multilingual-e5 dense retriever. Furthermore, LAURA integrates Tavily Search to access the latest legal information from external sources in real-time. This integrated approach demonstrates the significant potential of combining agentic reasoning with multi-source retrieval to create an accurate and interpretable legal question-answering system for the Vietnamese language.