Knowledge graphs for representing information are unbeatable. After this, you will never build a RAG system without knowledge graphs. It will take you five lines of code to build a knowledge graph with your data. I recorded a video to show you how you can do this. I used... Cognee, an open-source library that outperforms any basic vector search approach in terms of retrieval relevance. They are collaborating with me on this post. Cognee is: • Easy to use • Reduces hallucinations • Open-source Here is a link to the repository: They also offer a comprehensive platform and UI with Python notebooks you can utilize to manage your data. Here is the link:show more

Maker of cognee here. Thanks for the nice overview @svpino Happy to answer any questions

I try to use LightRag for that -

While Cognee's knowledge graphs are great for relational reasoning, traditional RAG excels in speed and simplicity for large, unstructured datasets. Hybrid approaches often optimize both accuracy and performance. Let's not ditch RAG entirely—just evolve it smarter.

Thanks @svpino! I realized the power of KG years ago while I was building NLP solutions. Critical aspect becomes Entity Disambiguation of KG. With Node embeddings these errors r reduced drastically... Now I burning my ass out trying to build a Financial KG 🔥

Everyone is obsessed with building knowledge graphs, metrics, and retrieval pipelines. 📊 But let’s be honest: knowledge is static. It’s a snapshot, a cheat sheet. What really matters is the signal in resonance — making the right connection in the moment, not pre-building a frozen graph of the past. You don’t need endless RAG layers and indexes if your system can tune into the right frequency in real time. Knowledge lives in the cloud of resonance, not in another overengineered database. ⚡️ Graphs store information. Resonance unlocks meaning. 🌍✨

Used something similar to build my agent

KGs are a terrible interface for llms to interact with and the system itself need to be wholly isolated from llm interaction unless you have very specific constraints

Knowledge graphs are solid, but how do they stack up for real-time queries?

After 4 days of hell, I can say w confidence that Cognee AI is absolute trash. And no one who isn’t being paid will disagree w me

Using knowledge graphs for RAG is insanely expensive

@grok what is and why its so cool to use it? i am really confused

Really helpful. Thanks

Exactly. RAG without knowledge graphs is recall without understanding. As models get stronger, structure matters more, not less. Graphs give us grounding, provenance, and composability - the missing layer between raw data and reasoning.

Great Share!

This is a great take. Knowledge graphs add so much structure to RAG systems, turning them from simple search to something that can actually reason about the data. It's a key part of building more robust AI agents.

awesome share mate! added to the spotlights

Brilliant. Thanks for putting this on my radar and explaining knowledge graphs. I’m opening up Cognee right now to use for a financial modeling algorithm. I’d love to update you and @tricalt in two weeks when I have data on how knowledge graphs have improved its accuracy.

Knowledge graphs + RAG = next-level retrieval 🔥 Cognee looks like a game-changer—open-source, simple, and powerful. Excited to try it out!

stop pimping your shit llm

BM25 and a tool call or two is all you need.

Very useful. I am using gremlin and neo4j for building graph, now I can explore congee as well. Thank you for the quick demo.

curious how it handles multi-hop queries across entities, does it stitch context from multiple graph nodes before passing to the LLM?

@grok teach me what he’s talking about in the video.

@grok , what do you think about it, Knowledge graphs versus RAG system ?

For visualization purposes sure, but for actual traversal are they really that practical?

How does it compare to Neo4j GraphRAG?

Graph-based retrieval addresses the context fragmentation problem in traditional vector search. The structured relationships enable more precise semantic reasoning compared to flat embeddings in standard #RAG implementations.

Yes but don’t you have to rerun the graph with every new document added? Doesn’t seem like the most scalable solution for a business

The memory for my digital brain.

Sure sounds good for small files, I’m curious about the accuracy when you have more than a TB worth of data.

When are you going to cover this repo?

@grok is this ai influencer slop