Input Hype:
“Neo4j Launches Breakthrough Infinigraph Architecture to Unify Transactional and Operational Workloads”
Output Hype Translation:
After over a decade of “clustering”, Neo4j finally figured out sharding, basically fixing the thing their clustering should’ve done from the start.
If you’re doing GraphRAG in the style of Microsoft’s original model , where the graph is inferred dynamically by the LLM and used primarily for retrieval or reasoning , then Neo4j is often overkill in my opinion. Not to mention its performance issues and toxic attitude to open source.
Lightweight, in-memory alternatives like NetworkX or even plain Python dictionaries are typically faster to set up, easier to integrate, and better suited to ephemeral, prompt driven graph reasoning.
Here are some other options :
NetworkX
LlamaIndex’s KnowledgeGraphIndex
LangChain with in-memory Python structures, Lets you model entities and relationships as dictionaries or lists, making it easy to traverse or query on demand.
Neo4j has bet the farm on GraphRag. Unfortunately - I don’t think the return on investment is there even if using Neo4j community edition - and even then - I think there are better options.
If you are using community edition - check out the DozerDB plugin which adds enterprise features to Neo4j community such as multi database. Its still in its infancy but has already implemented multi db and enterprise constraints. https://dozerdb.org
Will more information that could touch on what you mentioned be in their S-1 registration if they attempt an IPO? I need to search through all the court documents to see if there are any more details about prior years. Good thing I have LLMs to help with this now.
The analysis below seems to show that GraphRag with @Neo4j offers minimal improvement in context retrieval and only slight gains in answer relevancy and faithfulness.
Why would the investors of Neo4j take control of Neo4j Inc?
I believe the COO was removed as COO around the time I first heard this.
I hope Neo4j Inc does not get desperate and add terms to their Neo4j Community Edition making it proprietary or non-free like they did with their enterprise edition!
It seems from their repo (need to verify) that they did the right thing and changed the License to their own license name called Dgraph Community License (DCL)
Compare this to Neo4j, who added the commons clause to the AGPL License complete with FSF copyright and preamble. Neo4j kept the License as AGPL and reaped the benefits from open source community.
Neo4j’s approach made people think it was still open source under the AGPL, while dgraph’s approach was proper and clear.
Neo4j’s approach was deceitful in my opinion, and I believe it’s finally coming back to haunt them after some toxic rulings put the GPL structure at risk unless an upcoming appeal overturns it.
Good for dgraph for doing it right in the end and not making people think they were Apache licensed with commons.
I thought it was interesting that the 9th circuit made so many mistakes that they had to append the ruling. The fact that they got the defendants wrong is quite strange to me. The Graph Foundation was not part of the case any longer. They joined the case but then settled. The injunction itself that
Neo4j was very sloppy when adding the commons clause into the AGPL. They literally added the commons clause to the AGPL, leaving in the preamble and all, and call it the ‘Sweden License’. (Virtual name)
This ‘sweden license’ literally states it is a ‘free, copyleft license.’
copyleft = opensource according to Philip Rathle, Vice President of Products at Neo4j Inc. (6-ER-1368 ¶¶ 10-11 of appeal except of records).
Therefore according to Neo4j - it is free and open source.
Neo4j says their 'Sweden License' (AGPL + Commons) is both open source and not open source at the same time.
I call this: 'Schrödinger's Sweden License'
Keep an eye on the case - it is now in the second phase of the case where everyone will learn about Neo4j’s behavior. It would make a good Netflix mini series.
Input Hype: “Neo4j Launches Breakthrough Infinigraph Architecture to Unify Transactional and Operational Workloads”
Output Hype Translation: After over a decade of “clustering”, Neo4j finally figured out sharding, basically fixing the thing their clustering should’ve done from the start.