The rise of large language models (LLMs) has significantly transformed both the construction and application of information retrieval (IR) systems. Currently, IR and LLMs has limited interaction, where LLMs only serve as part of IR component and IR systems are separately constructed from LLMs. Such inter-independent architecture limits knowledge sharing and deep collaboration among LLMs and IR systems.In this paper, we introduce \emph{Self-Retrieval}, a novel end-to-end LLM-driven information retrieval architecture.Self-Retrieval integrates all the essential functions of IR systems into a single LLM, leveraging the inherent capabilities of LLMs throughout the IR process.Specifically, Self-Retrieval internalizes the retrieval corpus through self-supervised learning, transforms the retrieval process into a sequence of passage generation and assesses the relevance for reranking.Experimental results demonstrate that Self-Retrieval not only substantially outperforms previous retrieval approaches by a significant margin, but also can significantly boost the performance of LLM-driven downstream applications like retrieval augmented generation.