graphrag

Why we included this project

GraphRAG is the project Microsoft Research built to demonstrate the graph-based RAG approach it described in 2024, and it has become a common starting point for teams that need to reason over large private document collections. Rather than only retrieving chunks of text, its indexing pipeline uses LLMs to pull entities, relationships, and claims out of unstructured documents, then runs community detection to build hierarchical summaries the query engine can navigate. That structure lets the system answer global questions about an entire corpus, not just locate isolated passages, which is more than plain vector retrieval offers. Teams working with narrative-heavy material like internal wikis, research papers, or incident reports will find the CLI and Python API useful for building a knowledge-graph-backed search layer. One caveat: indexing is compute-intensive and the docs recommend prompt tuning for good results, so budget accordingly before running it at scale.