This Week in Neo4j: Memory, Knowledge Graph, GraphRAG, AI Agents and more

FEATURED COMMUNITY MEMBER: Adam Pingel

AGENT MEMORY: Inside Neo4j’s agent memory

Paul Iusztin takes a deep dive into neo4j-labs/agent-memory, explaining its three-tier memory model (short-term conversation, long-term typed entities, reasoning traces) and how the library’s three-tier graph architecture (short-term conversation, long-term typed entities, reasoning traces), POLE+O ontology, and SAME_AS deduplication pattern actually solve the agent memory problem.

KNOWLEDGE GRAPH: Sequential Systems. Simultaneous Crises.

Luanne Misquitta started a new series with a diagnosis: military, healthcare and emergency systems don’t have a headcount problem or a budget problem – they have a data architecture problem. When operational pressure arrives concurrently rather than sequentially, siloed HR systems can’t answer compound questions about who is available, qualified and safe to redeploy without breaking something else. That’s exactly the problem a connected, relationship-traversable data model is built to solve.

GRAPHRAG: How a Neo4j semantic layer makes your Text-to-SQL agent smarter and cheaper

Laurent Tande benchmarks two Text-to-SQL agents side by side: one feeding the full YAML schema to the LLM on every call, and the other querying a Neo4j semantic layer for only the relevant subgraph. Resulting in 20–30% fewer tokens on average, up to 10× less on simple queries, and ~10% better accuracy on complex multi-table joins. The repo is public and runnable in three minutes.

AI AGENTS: Build AI Agents that make better decisions on GCP with Neo4j

Michael Hunger covers the full stack for running production-grade agents on GCP with Neo4j: GraphRAG retrieval via hybrid Cypher queries, a semantic layer that helps agents navigate enterprise data sprawl, three-tier graph memory (short-term, long-term entities, reasoning traces) and context graphs that surface tacit decision patterns from agent traces. All of that with working code and integrations for ADK, Gemini Enterprise A2A and the Neo4j MCP server on Cloud Run.

HACKATHON: Aura Agents

The Aura Agent Hackathon is in full swing and projects can be submitted through June 15.  The community is already building some seriously impressive projects.
Complete the Building Agents in Neo4j Aura course by June 15, register for your $100 Aura credits and build an AI agent powered by a knowledge graph.  Prizes including Bose QuietComfort Ultra Headphones, a Raspberry Pi 5, Aura credits and a newly designed Neo4j “Agent” T-shirt.

POST OF THE WEEK: Yuval Shimon

Please share it if you like it!