I've been building Unize, a personal and collective knowledge management app, for the last three years with a small team. It works with essentially any type of data, from Notion-like blocks to named relationships between items. In 2022, it seemed like AI could be really helpful for users to help them work with knowledge and achieve goals, so I started building a bunch of AI systems related to that. One of those systems was to store unstructured data in the app as structured data, and another was to retrieve data in different formats.

Earlier this year, I started seeing the term "GraphRAG" floating around, which seemed pretty similar to what I had been working on in the background. I was surprised GraphRAG was taking off because apparently, building the graph is the hard part, and it doesn't seem like anyone has figured that out yet. I was talking to the CTO of Neo4j who was very enthused about GraphRAG, so I decided to convert our proof-of-concept into an API that anyone can use to generate a knowledge graph.

And that's what I'm releasing today! I still consider it really rough, but I created a benchmark for knowledge graph generation and it seems to outperform existing systems by substantial margins, so I'm releasing the alpha version of it ("Unize ST-0.5") today to get some feedback. It's pretty slow and expensive, but it's just a converted proof of concept. I think that human-level knowledge graph generation is very doable, it just requires a bunch of changes we haven't implemented yet, but we might if there's significant interest.

There are free API credits and a playground to generate and visualize graphs within the API dashboard, so it should be pretty easy to try out.

API: https://api.unize.org/signup

API Docs: https://developers.unize.org

Launch Post: https://blog.unize.org/p/introducing-unize-storage

22-Page Benchmark Report: https://developers.unize.org/kgstorage

Any thoughts and feedback would be greatly appreciated!