Beyond Context Graphs: Agentic Memory, Cognitive Processes, and Promise Graphs

If you're following my work, I have a couple of deep research pieces about agentic memory and the application of spacetime concepts. Right now, I'm also working on a book about agentic protocols. This particular book focuses on the question of better context for agents—following the ideas of context graphs and exploring how to actually build something beyond context graphs to manage decision traces.

The goal is to create an architecture for enterprise-level agents that follow rules, make their own decisions, and—most importantly—make explainable decisions. I'll also explore the ability for agents to learn and apply that learning based on past experience.

We'll talk extensively about concepts like agentic memory: why we need memory, how to build memory, and why memory is not just another RAG system. We'll cover how to apply decision traces, how they work, and why cognitive processes—just like memory structures—contribute to learning capabilities.

Beyond this, we'll explore promise theory and promise graphs as an extension of agentic action logs. This creates a rich trace from data signals to promises to actions, making the architecture multi-agent ready. We'll examine this in the scope of agent cooperation, where agents don't just act in isolation but coordinate through explicit promises and commitments.

My core assumption with this book is that we need to build sophisticated agentic memory that goes beyond context graphs, beyond knowledge graphs, and applies advanced topics like causality, temporal causality research, and a deep focus on time itself. This is extremely relevant to my memory book, but here the focus is specifically on agents—and how they work together.