Large language models (LLMs) are impressive, but they can hallucinate facts and struggle to keep their knowledge up-to-date. One solution is retrieval-augmented generation (RAG), which lets LLMs access external knowledge. But comparing different RAG algorithms fairly has been tough, and many open-source RAG tools are complex and hard to customize. Enter RAGLAB, a new modular framework for RAG research. It reproduces six existing algorithms and offers a standardized way to test and develop new ones. RAGLAB simplifies complex research by unifying core components and offering easy-to-use interfaces. Researchers can now easily swap different retrievers, generators, and instructions to see how they affect performance. The framework includes preprocessed Wikipedia databases and popular benchmarks, saving researchers valuable time and effort. RAGLAB's initial experiments, using several LLMs like Llama and GPT3.5, reveal fascinating insights. While some advanced RAG methods shone with larger models, they didn't show a clear advantage with smaller ones. Surprisingly, simpler RAG approaches often performed just as well. The research also confirms that LLMs struggle with multiple-choice questions, likely because added information confuses them. RAGLAB's user-friendly design makes it a powerful tool for both experts and newcomers to RAG research. A recent survey showed most users found it boosted their research efficiency significantly. As RAGLAB evolves, it promises to be a crucial driver of innovation in making LLMs smarter and more reliable.