Abstract: Against the dual background of the People-Centered City agenda and the deep digital transformation of territorial spatial planning, existing research has not yet adequately addressed the public need for precise, convenient, and low-cost intelligent retrieval of regulatory detailed planning information. This study first discusses a process-oriented pathway for cultivating planning agents, and then explains the construction logic of a planning information retrieval agent that integrates large language models (LLMs) and knowledge graphs within a retrieval-augmented generation (RAG) framework. On this basis, it proposes a three-stage method, using the Dify low-code development platform, for building ZoKnow, a public-participation-oriented regulatory planning retrieval agent. Seven statutory plans from Futian District, Shenzhen, including about 80,000 Chinese characters of planning text and 1,185 parcel-level control indicators, are used for application verification through prototype development, question- answering performance evaluation, and comparative analysis with the public version of the "One Map" platform. The results show that the proposed method is feasible and delivers superior retrieval performance. It is especially advantageous for terminology explanation, statutory-result traceability, and data analysis in public participation in regulatory planning. The study further discusses the potential of extending planning information retrieval agents toward multiple types of planning agents and their long- term significance for industry transformation. It provides a low-cost, reproducible, and scalable construction paradigm for planning agents that can support digital-intelligent planning transformation and technology-enabled spatial good governance.