Short-term and long-term memory together give an agent a working memory and a knowledge base — but they do not answer the hardest enterprise question: why did the agent decide that? This module introduces reasoning memory, the layer that captures every tool call, intermediate thought, and causal chain, turning your agent’s decision history into a queryable institutional knowledge base — a context graph.
By the end of this module, you will:
-
Explain what a context graph is and how reasoning memory implements the "context" layer
-
Instrument a Pydantic AI agent to record complete reasoning traces into Neo4j
-
Write Cypher queries that traverse from a tool call back through reasoning steps to the originating entity
-
Use vector similarity to find past traces for similar tasks (agent learning from experience)
-
Produce a human-readable audit report of an agent’s decisions from the graph