all-agentic-architectures

Why we included this project

Teams trying to choose between agent patterns often end up wading through scattered blog posts and half-finished demos. This project packages 35 well-known architectures, including Reflexion, LATS, GraphRAG, MemGPT, Voyager, ReAct, and BrowserAgent, as runnable Python classes behind one uniform .run(task) interface and a single return shape. Because every pattern shares that contract, you can move from a self-critique loop to a tree-search or retrieval-grounded design without touching your downstream code. It is built on LangGraph and supports nine LLM providers, so it slots into whatever model stack you already use. Each pattern comes with a fully executed notebook, and a 17-task benchmark ranks the architectures against one another, which beats trusting vibes when you are deciding which design to take to production.