# GraphRAG
Using GraphRAG approach, we can make LLMs capable of answering more accurately from private data, along with references to the data points that support the answer.
This approach generally works better than simple semantic (vector) search by overcoming its inability to retrive enough relevant context for the questions in most cases.
The approach involves using LLMs to analyse the given private data, generate metadata about each data point, build graph of communities from the generated metadata, generate metadata about each community, and build a graph of communities of communities and so on.
This helps LLMs to gather relevant context faster and more accurately, by searching the metadata to find relevant data points faster, and sometimes even provide accurate answers that are not directly present in the data, but was discovered during graph generation.