(LSJ) Context engineering advanced

/

Paper: A Survey of Context Engineering for Large Language Models

  • Authors: Lingrui Mei, Jiayu Yao, Yuyao Ge, Yiwei Wang, Baolong Bi, Yujun Cai, Jiazhi Liu, Mingyu Li, Zhong-Zhi Li, Duzhen Zhang, Chenlin Zhou, Jiayi Mao, Tianze Xia, Jiafeng Guo, and Shenghua Liu.

  • Source: arXiv e-print (Preprint)

  • Date: July 2025

Reference List


Journal Articles & Preprints

  1. Mei, L., Yao, J., Ge, Y., et al. (2025) A Survey of Context Engineering for Large Language Models. arXiv preprint [2507.13334]. Available at: https://arxiv.org/abs/2507.13334 (Accessed: 4 November 2025).

  2. Lewis, P., Oğuz, B., Yarats, D., et al. (2020) ‘Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks’. In: Advances in Neural Information Processing Systems 33. Vancouver, Canada: NeurIPS.

  3. Brown, T. B., Mann, B., Ryder, N., et al. (2020) ‘Language Models are Few-Shot Learners’. In: Advances in Neural Information Processing Systems 33. Vancouver, Canada: NeurIPS.

Technical Reports & Online Guides

  1. Anthropic (2025) Effective context engineering for AI agents. Available at: https://www.anthropic.com/engineering/effective-context-engineering-for-ai-agents (Accessed: 4 November 2025).

  2. LlamaIndex (2025) Context Engineering – What it is, and techniques to consider. Available at: https://www.llamaindex.ai/blog/context-engineering-what-it-is-and-techniques-to-consider (Accessed: 4 November 2025).


2. Notable Products and Frameworks in the Market


Context Engineering is primarily implemented using open-source frameworks and vendor platforms that specialize in building LLM applications, especially those relying on RAG and Agent architectures.

LangChain is a Development Framework

LangChain Provides a cohesive framework for building agentic workflows and RAG pipelines. It offers modules for memory management, tool calling, and chaining LLM steps, which are all fundamental to context orchestration.

LlamaIndex is Data Framework

LlamaIndex Specializes in connecting LLMs to external data sources. It is core to the Retrieval component of context engineering, offering various indexing and query transformation strategies to retrieve the most relevant information for the context window.

LlamaStack is Integrated Ecosystem

A framework and ecosystem built to specifically simplify the creation and deployment of applications using Meta's Llama models. Focuses on optimized data processing, training, and streamlined inference/deployment for Llama-based context.

Weaviate / Pinecone are Vector Databases

These are the storage backends for long-term memory and external knowledge. They allow for semantic search, enabling the Retrieval-Augmented Generation (RAG) process that feeds precise, context-specific information to the LLM.

Zep is Context Management Platform

A dedicated platform for managing complex conversational memory and state. It specifically addresses challenges related to session history, user profiles, and filtering context for long-running AI applications.

AutoGen is Multi-Agent Framework

Developed by Microsoft, it focuses on building multi-agent systems where different AI agents collaborate to solve complex tasks. Context engineering here involves defining communication protocols and isolating context for each sub-agent.

Microsoft Agent Framework is Agent Orchestration Framework

A comprehensive, open-source SDK combining the strengths of AutoGen and Semantic Kernel. It features graph-based workflows, robust state/context management, and high control over multi-agent execution paths.

Microsoft Agent Lightning is Optimization Framework

A cutting-edge framework that enables Reinforcement Learning (RL)-based training of any existing AI agent (e.g., those built with LangChain, AutoGen) by decoupling training from execution. Focuses on optimizing agent behavior and long-horizon context logic.

Google (Vertex AI & Gemini)Platform / Model Suite

Emphasizes the use of massive context windows (e.g., 1M+ tokens in Gemini models) to directly handle large amounts of multimodal context (text, video, audio) in one call. Vertex AI Agent Builder provides enterprise tools for designing and deploying agents.

Meta (Research Focus)

Research / Open SourceMeta's approach often centers on optimizing their Llama family of models for efficiency and context handling. Research highlights include frameworks for autonomous context auditing and continuous learning where an agent enriches conversation history with structured metadata/tags to manage its own knowledge base.

LangGraph Graph-based Orchestration

An extension of LangChain that allows for stateful, cyclical workflows using graphs. It provides fine-grained control over which components (RAG, memory, tool calls) are executed and what context is passed at each step of a complex multi-agent process.