Building Context Graphs for AI Agents, Will Lyon, Neo4j

A context graph is a knowledge graph designed to capture not just data and events, but the “why” behind decisions—the missing layer in many AI and enterprise systems. Instead of simply logging what happened (like a rejected transaction), it connects entities (people, accounts), events (transactions), and context (policies, decisions, precedents) to explain how and why an outcome occurred. The challenge is that this reasoning is often fragmented across multiple systems, difficult to model, and complex to query, especially as decisions influence future decisions. By unifying all of this into a single graph, context graphs enable traceability, auditability, and trust, turning opaque decision-making—whether from AI agents or business systems—into something transparent and explainable.

0:00 What is a context graph and why does it matter for AI agents?
3:00 Why the “missing why” is the central problem in enterprise AI decision-making
6:30 Live demo: context graph AI agent for credit approval in financial services
11:00 How the agent traverses the graph using hybrid vector and graph search
13:00 Building context graphs with the Neo4j agent memory Python package
16:00 Short-term, long-term, and reasoning memory: three abstractions explained
18:30 Agent framework integrations: Google ADK, AWS Strands, Microsoft