Context Engineering for Multi-Agent Systems
A collection of runnable notebooks implementing a domain-agnostic Context Engine, a transparent architecture for orchestrating multi-agent systems through semantic context and reasoning rather than hard-coded workflow logic.
This repository is a production-oriented blueprint for building universal, domain-agnostic multi-agent systems through high-level semantic orchestration instead of hand-coded workflow branches, aiming to keep the resulting system's reasoning transparent and observable. It is organized as a series of notebook chapters building the Context Engine up piece by piece, with several alternate editions: one built with GPT-6 Astra's computer-use capability for perceiving and acting in interactive environments, one implemented purely as markdown protocol files running inside a Claude Project with no code or vector database, one built on NVIDIA NIM Nemotron with a real-time DAG planner and Docker or Railway deployment, and one built on a LangChain substrate.
A changelog tracks updates across editions, and a Gradio-based standalone UI notebook can be deployed with one command to Hugging Face Spaces or run with a public Colab link.