**Abstract:** Retrieval-augmented generation (RAG) has significantly improved language understanding systems, but challenges such as poor retrieval quality for complex questions, inefficiencies in knowledge re-retrieval, and lack of personalized responses persist. To address these issues, the authors introduce ERAGent, an advanced framework that enhances RAG. ERAGent introduces an Enhanced Question Rewriter and Knowledge Filter module to improve retrieval quality, a Retrieval Trigger to curtail extraneous external knowledge retrieval, and a Personalized LLM Reader to tailor responses to individual user profiles. The Experiential Learner module enables ERAGent to learn user profiles from interactions, enhancing efficiency and personalization. Evaluations across six datasets and three question-answering tasks demonstrate ERAGent's superior accuracy, efficiency, and personalization, highlighting its potential to advance RAG and practical applications. **Introduction:** Large Language Models (LLMs) have made significant strides in generalization and adaptability, but challenges like hallucinations, temporal misalignments, and context processing issues remain. RAG integrates LLMs with external knowledge sources to improve response accuracy. The basic RAG architecture consists of a retriever and a read module, but it faces limitations in retrieval quality and response reliability. Advanced RAG modules, such as the question rewriter and post-reading fact-checking, have been developed to enhance these aspects. ERAGent addresses these challenges by integrating an Enhanced Question Rewriter, Retrieval Trigger, Knowledge Filter, Personalized LLM Reader, and Experiential Learner module. **Methodology:** ERAGent's pipeline includes: - **Enhanced Question Rewriter:** Refines questions into standardized and fine-grained queries. - **Retrieval Trigger:** Evaluates if a query exceeds the AI assistant's knowledge boundary. - **Knowledge Retriever:** Retrieves external knowledge. - **Knowledge Filter:** Refines retrieved knowledge by filtering out irrelevant context. - **Personalized LLM Reader:** Generates responses tailored to individual user contexts. - **Experiential Learner:** Learns user profiles from interactions, updating the knowledge base and user profile. **Experiments:** - **One-Round Open-Domain QA Task:** ERAGent's modules improve response accuracy and quality. - **One-Round Multi-Hop Reasoning QA Task:** The synergistic effect of the Enhanced Question Rewriter and Knowledge Filter is effective in complex logic reasoning tasks. - **Multi-Session Multi-Round QA:** ERAGent provides personalized responses and enhances response efficiency without compromising quality. **Conclusion:** ERAGent significantly improves RAG by enhancing question rewriting, response robustness, and personalization. Comprehensive evaluations across diverse datasets and tasks demonstrate its superior performance, making it a promising solution for practical applications in AI-based assistants.**Abstract:** Retrieval-augmented generation (RAG) has significantly improved language understanding systems, but challenges such as poor retrieval quality for complex questions, inefficiencies in knowledge re-retrieval, and lack of personalized responses persist. To address these issues, the authors introduce ERAGent, an advanced framework that enhances RAG. ERAGent introduces an Enhanced Question Rewriter and Knowledge Filter module to improve retrieval quality, a Retrieval Trigger to curtail extraneous external knowledge retrieval, and a Personalized LLM Reader to tailor responses to individual user profiles. The Experiential Learner module enables ERAGent to learn user profiles from interactions, enhancing efficiency and personalization. Evaluations across six datasets and three question-answering tasks demonstrate ERAGent's superior accuracy, efficiency, and personalization, highlighting its potential to advance RAG and practical applications. **Introduction:** Large Language Models (LLMs) have made significant strides in generalization and adaptability, but challenges like hallucinations, temporal misalignments, and context processing issues remain. RAG integrates LLMs with external knowledge sources to improve response accuracy. The basic RAG architecture consists of a retriever and a read module, but it faces limitations in retrieval quality and response reliability. Advanced RAG modules, such as the question rewriter and post-reading fact-checking, have been developed to enhance these aspects. ERAGent addresses these challenges by integrating an Enhanced Question Rewriter, Retrieval Trigger, Knowledge Filter, Personalized LLM Reader, and Experiential Learner module. **Methodology:** ERAGent's pipeline includes: - **Enhanced Question Rewriter:** Refines questions into standardized and fine-grained queries. - **Retrieval Trigger:** Evaluates if a query exceeds the AI assistant's knowledge boundary. - **Knowledge Retriever:** Retrieves external knowledge. - **Knowledge Filter:** Refines retrieved knowledge by filtering out irrelevant context. - **Personalized LLM Reader:** Generates responses tailored to individual user contexts. - **Experiential Learner:** Learns user profiles from interactions, updating the knowledge base and user profile. **Experiments:** - **One-Round Open-Domain QA Task:** ERAGent's modules improve response accuracy and quality. - **One-Round Multi-Hop Reasoning QA Task:** The synergistic effect of the Enhanced Question Rewriter and Knowledge Filter is effective in complex logic reasoning tasks. - **Multi-Session Multi-Round QA:** ERAGent provides personalized responses and enhances response efficiency without compromising quality. **Conclusion:** ERAGent significantly improves RAG by enhancing question rewriting, response robustness, and personalization. Comprehensive evaluations across diverse datasets and tasks demonstrate its superior performance, making it a promising solution for practical applications in AI-based assistants.