The Three Hidden Graphs in Every AI Agent

Presentation Deck (PDF)

Download the complete presentation slides

Knowledge Graph demo: RAG vs Graph-RAG

300 hotel FAQs. The graph computes AVG and COUNT and returns an honest 'no results', where vector RAG fabricates statistics.

If you find it useful, give it a ⭐!

Graph Memory + Reasoning Memory demos

Demo 03 (graph memory, multi-hop) and Demo 06 (reasoning memory, reverse audit over :TOUCHED). MIT-0.

If you find it useful, give it a ⭐!

Blog series: Stop AI Agents from Losing Memory

The write-ups behind the graph memory and reasoning memory demos

Strands Agents SDK

Build production-ready agents in a few lines. Tools and hooks power the demos: the recorder is a HookProvider on the events the agent already emits.

neo4j-graphrag (Python)

Builds the knowledge graph and graph memory: an LLM extracts entities and relationships against a pinned schema

Neo4j agent-memory (reasoning traces)

The official SDK that records the decision traces and the :TOUCHED edges the reverse audit walks

What is GraphRAG? (Neo4j)

A knowledge graph retrieves grounded, traceable context instead of guessing