The legality review of tender documents is a critical process in procurement projects, ensuring fairness and compliance in the bidding process. However, existing automatic review methods cannot deeply understand complex document statements, and constantly changing legal and regulatory data also brings difficulties to building models. To address these challenges, we propose a Retrieval-Augmented-Generation-based framework for automatic legality review, leveraging large language models (LLMs) with enhanced semantic understanding. Our method involves constructing a knowledge base, using vector similarity to extract relevant clauses, and proposing an RAG pipeline that retrieves pertinent cases to prompt LLMs for legality assessment. Experimental results show that our approach achieves high precision and recall in legality classification, effectively identifying critical clauses while ensuring compliance. The results demonstrate our method is effective and accurate, offering a practical solution for tender document reviews. Improving both the speed and reliability of automatic tender document reviews.