A Survey of Context Engineering for Large Language Models

Abstract: The performance of Large Language Models (LLMs) is fundamentally determined by the contextual information provided during inference. This survey introduces Context Engineering, a formal discipline that transcends simple prompt design to encompass the systematic optimization of information payloads for LLMs. We present a comprehensive taxonomy decomposing Context Engineering into its foundational Components and the sophisticated Implementations that integrate them into intelligent systems. We first examine the foundational Components: (1) Context Retrieval and Generation, encompassing prompt-based generation and external knowledge acquisition; (2) Context Processing, addressing long sequence processing, self-refinement, and structured information integration; and (3) Context Management, covering memory hierarchies, compression, and optimization. We then explore how these components are architecturally integrated to create sophisticated System Implementations: (1) Retrieval-Augmented Generation (RAG), including modular, agentic, and graph-enhanced architectures; (2) Memory Systems, enabling persistent interactions; (3) Tool-Integrated Reasoning, for function calling and environmental interaction; and (4) Multi-Agent Systems, coordinating communication and orchestration.

Introduction. The advent of LLMs has marked a paradigm shift in artificial intelligence, demonstrating unprecedented capabilities in natural language understanding, generation, and reasoning [103, 1059, 453]. However, the performance and efficacy of these models are fundamentally governed by the context they receive. This context—ranging from simple instructional prompts to sophisticated external knowledge bases—serves as the primary mechanism through which their behavior is steered, their knowledge is augmented, and their capabilities are unleashed. As LLMs have evolved from basic instruction-following systems into the core reasoning engines of complex applications, the methods for designing and managing their informational payloads have correspondingly evolved into the formal discipline of Context Engineering [25, 1256, 1060]. The landscape of context engineering has expanded at an explosive rate, resulting in a proliferation of specialized yet fragmented research domains. We conceptualize this landscape as being composed of foundational components and their subsequent implementations.

Discussion / Conclusion. This survey has presented the first comprehensive examination of Context Engineering as a formal discipline that systematically designs, optimizes, and manages information payloads for LLMs. Through our analysis of over 1400 research papers, we have established Context Engineering as a critical foundation for developing sophisticated AI systems that effectively integrate external knowledge, maintain persistent memory, and interact dynamically with complex environments. Through this systematic examination, we have identified several key insights. First, we observe a fundamental asymmetry between LLMs’ remarkable capabilities in understanding complex contexts and their limitations in generating equally sophisticated outputs. This comprehension-generation gap represents one of the most critical challenges facing the field.