This repository provides a comprehensive survey of Agentic Retrieval-Augmented Generation (RAG) systems, exploring how AI agents enhance traditional RAG pipelines. It targets researchers and practitioners interested in advanced AI architectures for complex reasoning, multi-domain knowledge retrieval, and document-centric workflows, offering a taxonomy, comparative analysis, and practical implementation details.

How It Works

Agentic RAG integrates autonomous AI agents into the RAG process, leveraging core patterns like reflection, planning, tool use, and multi-agent collaboration. This approach allows agents to dynamically adapt to task requirements, perform multi-step reasoning, and interact with external tools, overcoming the limitations of traditional RAG systems in handling complex, dynamic, and orchestrated workflows.

Quick Start & Requirements

• Install/Run: The repository primarily serves as a survey and resource hub. Specific implementation examples are provided via linked notebooks.
• Prerequisites: Notebooks showcase usage with libraries like LangChain, LlamaIndex, LangGraph, AutoGen, CrewAI, and vector databases (Chroma, FAISS, Weaviate, Redis). Cloud platforms like Google Cloud, AWS, and IBM Watsonx.ai are also featured.
• Resources: Setup complexity varies by notebook, potentially requiring API keys, specific Python versions, and significant computational resources for LLM interactions.
• Links: AgenticRAG-Survey GitHub, DeepLearning.AI, Weaviate Blog, LangGraph Tutorials

Highlighted Details

• Taxonomy: Classifies Agentic RAG systems into categories like Single-Agent, Multi-Agent, Hierarchical, Corrective, Adaptive, and Graph-Based RAG.
• Workflow Patterns: Details strategies such as Prompt Chaining, Routing, Parallelization, Orchestrator-Workers, and Evaluator-Optimizer.
• Comparative Analysis: Contrasts Traditional RAG, Agentic RAG, and Agentic Document Workflows (ADW) across features like context maintenance, adaptability, and scalability.
• Applications: Showcases use cases in healthcare, education, legal analysis, finance, and customer support.

Maintenance & Community

The survey paper is authored by Aditi Singh, Abul Ehtesham, Saket Kumar, and Tala Talaei Khoei. The repository is a static survey, with updates noted on the paper and images.

Licensing & Compatibility

The repository itself does not specify a license. The underlying code examples in linked notebooks will adhere to the licenses of their respective libraries (e.g., MIT for LangChain, Apache 2.0 for LlamaIndex).

Limitations & Caveats

This repository is a survey and does not provide a single, runnable framework. Implementation details and dependencies vary significantly across the linked examples, requiring users to consult individual notebook requirements. Challenges like coordination complexity in multi-agent systems and ethical AI considerations are highlighted as ongoing research areas.