About

Context Patterns is a pattern catalog and guide collection for context engineering, the discipline of designing what information AI systems work with at inference time.

The site exists because the same problems keep showing up across teams building LLM applications: context windows that fill up with noise, agents that degrade after a few turns, RAG pipelines that retrieve documents the model ignores, multi-agent systems where every agent drowns in context meant for other agents. These are engineering problems, and engineering problems have recurring solutions.

What's Here

The pattern catalog documents those solutions. Each pattern describes a specific context engineering problem, explains the mechanism behind the solution, and shows when to use it and when not to. The patterns are opinionated; they take positions on what works and say so directly.

The guides apply those patterns to specific domains: RAG pipelines, coding agents, code generation, multi-agent frameworks. They assume you're building LLM applications professionally and are past "how do I call the API." If you're looking for an introduction to the discipline itself, start with the practitioner's guide.

The research page collects the papers, benchmarks, and practitioner writing that the patterns draw from, annotated with why each source matters and what it contributed. Context Lens is an open-source tool that lets you see what your AI actually sees by intercepting LLM API calls and visualizing context window composition.

Context Patterns is built and maintained by Lars de Ridder.