This Week in Neo4j: NODES, relate, GraphRAG, Agentic AI and more

Photo of Alexander Erdl

Alexander Erdl

Staff Community Manager

Emmanuela Opurum

Welcome to This Week in Neo4j, your fix for news from the world of graph databases!

NODES 2026 just happened last week – and the first videos are already in our playlist. During the event, we announced Relate, a new open-source graph database that runs in-process/in-memory and speaks Cypher. For the launch, we are hosting a Hackathon (submit by December 11).

Also this week: have fun with GrapHRAG in a playable movie quiz and a guide to giving a Pydantic AI agent a graph to work with and remember.

Happy Graphing,

Alexander Erdl

PS: Join our Neo4j User Research panel! Sign up here to share your experiences with a researcher and influence the future of Neo4j products.What’s in it for you? A chance to connect directly with product development teams, get paid compensation, hear about what we are working on, and more!

COMING UP!

 

Emmanuela designs scalable, secure multi-cloud solutions (AWS, Azure, GCP) that turn complex business requirements into practical results, and mentors emerging tech talent through Code Your Future.

Connect with her on LinkedIn.

She was speaking at NODES 2026: “From Vector Search to GraphRAG: Architecting AI Systems That Actually Work in Production “, where she compared vector search and GraphRAG, covering knowledge graphs, retrieval orchestration, LLM integration, observability, and production strategies for accurate, explainable enterprise AI.


Emmanuela Opurum

NODES: NODES 2026 Wrap-Up and Call for Papers for NODES AI 2027


Last week, we finished NODES 2026, and with so many great sessions during the event, it was hard to pick what to watch, so we have a few sessions already in our NODES YouTube Playlist before the rest goes up in a few weeks. Particularly, the chat between Andrew Ng and Emil Eifrem is one to put on your “Watch Later” list.
During the event, we also announced a few things (see the post below) and opened the Call for Papers for NODES AI, happening in April 2027.  Please submit!

 

EMEDDED GRAPH DATABASE: Meet Relate by Neo4j: An Embedded Database for AI Agents


We are happy to announce Relate: a new open-source graph database that runs in-process/in-memory, with no server to set up, and it speaks Cypher, so agents can store what they learn as connected facts rather than loose text. In the demo, Claude Code audits the 24 direct dependencies of Google’s Agent Development Kit by building one small throwaway graph per package, where each package is a node and each “needs” link is a relationship.
Learn more on relatedb.ai and join our Hackthon for your chance to win a Mac Studio!

 

GRAPHRAG: GraphRAG – The Online Movie Quiz


Christoffer Bergmann turns his GraphRAG card game into a playable web quiz that shows the difference between plain RAG and GraphRAG. Players act as the LLM and can use a RAG lifeline (the top five cast of a movie) or a GraphRAG lifeline (the full path between movie and actor). All questions are generated on the fly from a Neo4j Aura graph built from Wikidata, which gives over half a million possible questions. The post also covers the Cypher side: one query shape takes minutes, while moving the uniqueness check after a random ordering returns an answer in seconds. Source code is on GitHub.

 

AGENTIC AI: Giving a Pydantic AI Agent a Graph to Think and Work with


Kaustubh Darekar shows how to connect a Pydantic AI agent to Neo4j in two steps: first through the official Neo4j MCP server, plus typed custom tools, schema-validated answers, and human approval before risky queries run; then through neo4j-graphrag retrievers, where one Cypher retrieval query turns each matched text chunk into a dated passage linked to the companies it mentions. Persistent memory is also stored in the graph (via neo4j-agent-memory), so you can query, audit, and delete what the agent knows. Both parts come as notebooks in the neo4j-agent-integrations repo.

 

STARTUPS: Seismora

Vito Palermo helped build operating systems and hardware for the early internet, and he sees AI creating the same kind of distributed systems problem all over again. His stealth startup, Seismora, co-founded with Dino Farinacci, is an intelligent control plane that routes AI workloads across any network, neocloud, or edge device based on latency, sovereignty, and cost. Palermo calls it cognitive routing. He’s already living a small version of it at home, where a $300 camera and single-board computer spot the coyotes in his backyard and ask a hosted model what kind they are. Watch the interview

Are you building on graphs? The Neo4j Startup Program gives Series B and earlier startups up to $16K in Aura credits, time with graph engineers, and co-marketing, such as the interviews above. Apply now

seismora

 

 

POST OF THE WEEK: Miodrag Cekikj

Please share it if you like it!