Replies: 1 comment
-
Dify doesn't natively support Graph RAG (retrieval over a knowledge graph or multi-hop/entity-relation traversal). Its workflow engine is graph-based in the sense that you can build complex, multi-step pipelines with nodes for retrieval, agents, and generation, orchestrated as a directed graph—see the GraphEngine implementation. However, this is about workflow orchestration, not graph-structured knowledge retrieval. The RAG pipeline in Dify supports modular retrieval (keyword, semantic, hybrid, reranking) and multi-dataset routing, but does not implement multi-hop or graph-based retrieval over a knowledge graph out of the box (KnowledgeRetrievalNode, RetrievalService). The multi-dataset routers can dynamically select among retrieval tools, which could be extended for more complex strategies, but they don't provide knowledge graph traversal or multi-hop reasoning natively. If you need true Graph RAG, you would need to extend Dify with custom nodes or plugins that implement graph-based retrieval logic, or integrate with an external system that provides this capability. The current architecture is flexible and could support such extensions, but it is not available out of the box. To reply, just mention @dosu. How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
Self Checks
Content
can i construct a Graph RAG with Dify?
Beta Was this translation helpful? Give feedback.
All reactions