The exponential rise of wireless devices is creating unprecedented spectrum demand, forcing spectrum management to make finer decisions across space, time, and device constraints. Policymakers and engineers must cope with massive heterogeneous data—from policy proceedings, legal regulations, to license databases—often scattered as text or tables, requiring substantial manual effort for integration, search, and interpretation. Most of this information is formatted for human consumption and is not readily machine‑readable.\ \ To tackle these challenges, we introduce SpecMind, a Multi‑Agent Retrieval‑Augmented Generation (RAG) system for spectrum intelligence. The architecture consists of a central agent that orchestrates specialized sub‑agents; each sub‑agent retrieves relevant excerpts from policy documents, legal texts, or license repositories and feeds them back for unified reasoning and answer generation.\ \ We also release SpecBench, a Q&A dataset built on real‑world license records and policy proceedings, filling the evaluation gap for RAG systems in the spectrum domain. Experiments show SpecMind outperforms generic RAG baselines across diverse spectrum queries, achieving over an 80% win rate. The agent‑based design improves retrieval precision, contextual reasoning, and overall task completion.\ \ Review: SpecMind demonstrates how coordinated multi‑agent retrieval can unlock intelligent spectrum management in environments where structured and unstructured data coexist.