What does it mean for an AI agent to think in graphs? In this session, we’ll look inside the mind of a graph-native detective agent that investigates a Neo4j crime graph powered by Microsoft Foundry.

Using the POLE model (People, Objects, Locations, Events) hosted on Aura, we’ll follow Holmsworth Marot, a Golden Age–style detective persona, as it connects to the graph through the Model Context Protocol and the Microsoft Agent Framework. The agent uses Cypher to traverse suspects, victims, locations, and events – and then explains its findings in the voice of a Golden Age detective, turning graph queries into narrative crime investigations.

We’ll walk through how the crime graph is modeled, how Cypher tools are exposed to the agent, and how we instrument the agent to reveal the sequence of graph queries and subgraphs it uses while investigating a case. You’ll see an ‘investigation trace’ that shows which Cypher queries were executed, which parts of the Neo4j graph were inspected, and how those pieces are combined into a narrative explanation.