Are Knowledge Graphs in RAG better than regular vector RAG?

Are Knowledge Graphs in RAG better than regular vector RAG? I’ve been asked 3x this week and tried my best to simplify the answer:

For most use-cases, it’s not worth. Avoid the added complexity and iterate on your prompts first.

That said, knowledge graphs can add value when:

1. The relationships themselves between the documents matter as much as the content itself
2. You care about 2nd or 3rd-degree document connections
3. The relationships are fuzzy, evolving, or hard to formalize

This is a simplified answer. Reach out and I can tell you more!

Introducing ANYTHING graph-related into tech increases the complexity, therefore cost, of your solution

There are some RAG systems which can “automatically” sort and update the document connections when you vectorize them. Most likely these connections won’t be high-quality / informative and you will just end up paying for extra complexity.

If you already know which relationships matter ahead of time, you’re probably better off with a well-designed relational database and letting the agent query it intelligently 🙂

If you’ve used them and seen better results, or found a use-case I haven’t mentioned, feel free to reach out :)