Context Engineering

Context engineering can be defined as the practice of designing systems that provide a Large Language Model (LLM) and AI agents with all the necessary information to complete a task effectively. It goes beyond prompt engineering since it focuses on building a comprehensive and structured context from various sources like instructions, external knowledge, memory, tools, and state. The central idea is that the success of a complex LLM-based system depends more on the quality and completeness of the context provided than on the specific wording of the prompt itself.

Tobi Lütke, the CEO of Shopify, coined the term context engineering in a tweet on June 19, 2025. He defined context engineering as the art of providing all the context for the task to be plausibly solvable by the LLM. This novel concept captures the essence of the current evolution of LLM-based systems, inspiring others (like me) to understand and define this emerging discipline. Since then, I've been working on a book entitled Context Engineering: Build Consistent, Accurate, Predictable AI Systems, published by Manning. Currently, the early access version of the book is available, and you can get it here.

This GitHub repository is intended to be a companion resource for this book and a reference for practitioners looking to understand and adopt the context engineering principles.

Table of contents

This book aims to provide a strong, general-purpose theoretical foundation for context engineering, supported by hands-on examples. Its table of contents is the following:

1. Introduction to context engineering
2. Instructions for AI agents
3. External knowledge and retrieval
4. Tools in AI agents
5. Memory and state in agentic systems
6. User prompts for LLMs
7. Context management and orchestration
8. Evaluation and observability
9. Governance and operations
10. AI frameworks for context engineering
11. Context engineering for software development
12. The state of the art in context engineering
    Appendix A. The AI ecosystem
    Appendix B. References and further reading

Each chapter of this book begins by explaining the underlying principles and patterns of each thematic block. The final part of each chapter then presents specific examples, all of which are available in this GitHub repository. New examples will continue to be added, and existing ones maintained, even after the book is published. The goal is to provide an open-source, up-to-date reference for everyone interested in context engineering.

Examples

This repository organizes examples by chapter to help you explore context engineering concepts in practice.

Chapter 1. Introduction to context engineering

This chapter provides the foundations for interacting with different model providers:

• Basic interaction (OpenAI): Python · Jupyter · JavaScript · Java
• Basic interaction (Anthropic): Python · Jupyter · JavaScript · Java
• Basic interaction (Google): Python · Jupyter · JavaScript · Java
• Basic interaction (Ollama): Python · Jupyter · JavaScript · Java

Chapter 2. Instructions in AI agents

This chapter covers the definition and usage of instructions (system prompts, agent skills, instructions artifacts) as a foundation layer to shape the model behavior:

• System prompts (OpenAI): Python · Jupyter · JavaScript · Java
• System prompts (Anthropic): Python · Jupyter · JavaScript · Java
• System prompts (Google): Python · Jupyter · JavaScript · Java
• System prompts (Ollama): Python · Jupyter · JavaScript · Java
• Agent skills: project-notetaker
• Instruction artifacts: task-tracker

Chapter 3. External knowledge and retrieval

This chapter explores different patterns for providing external knowledge to a model:

• Retrieval-Augmented Generation (RAG): Python · Jupyter
• Advanced RAG: LangChain · Qdrant · PageIndex · RAGFlow
• Context stuffing: Python · Jupyter
• Cache-Augmented Generation (CAG): Python · Jupyter

Chapter 4. Tools in AI agents

This chapter focuses on extending the capabilities of AI agents through tools:

• Function calling: Python · JavaScript · Java
• Command-Line Interface (CLI): workspace-analyzer
• Model Context Protocol (MCP) server: Python · JavaScript · Java

Chapter 5. Memory and state in agentic systems

This chapter explores how to maintain information across interactions using memory and state:

• Session memory: Python
• Long-term memory (Mem0): Python · JavaScript · Java
• Long-term memory (Cognee): Python
• Memory coach: Python
• Session state: Python · JavaScript · Java
• Workflow state: Python · JavaScript · Java

Chapter 6. User prompts for LLMs

This chapter focuses on the design and optimization of user prompts, including techniques like few-shot prompting, prompt chaining, Chain-of-Thought, or ReAct:

• Few-shot prompting: Python · Jupyter · JavaScript · Java
• Prompt chaining: Python · Jupyter · JavaScript · Java
• Chain-of-Thought vs ReAct (DSPy): Python · Jupyter

Chapter 7. Context management and orchestration

This chapter focuses on managing and orchestrating context in complex agentic systems, including context compression, hierarchical context, or multi-agent patterns:

• Context compression: LLMLingua
• Filesystem context: OpenViking
• Collaborative agents: CrewAI
• Multi-agent router: LangGraph
• Orchestration: DeepAgents
• Agent-to-Agent (A2A): Python

Chapter 8. Evaluation and observability

This chapter covers evaluation and observability for context-aware systems:

• Metrics: DeepEval
• Evals: Promptfoo
• LLM-as-judge: Ragas
• Observability: Langfuse
• Observability: LangSmith

Chapter 9. Governance and operations

This chapter covers governance, human oversight, and operational patterns for context-aware systems:

• Personally Identifiable Information (PII) redaction: Microsoft Presidio
• Output validation: Pydantic
• Bias detection: Fairlearn
• AI gateway: [LiteLLM](./ch09/litellm_g