One API for different knowledge graphs in RAG
A Reddit user said that a knowledge graph in RAG can mean several different things. They listed RDF/triple store, property graph, networkx graph, and agent-memory graph as examples. They said they are building one Python API across these graph types so retrieval and traversal logic can stay the same when the backend changes.
Key points
- The post says knowledge graph means different things in different RAG projects.
- The author says switching backend often forces teams to rebuild retrieval and traversal code.
- They are trying to place one Python API across several graph families.
- The goal is to keep the search path the same while changing the storage setup by config.
- The author asks whether a shared abstraction is useful or too broad for graph RAG.
Quick term guide
- knowledge graph
- A structured map that links facts, topics, or ideas and shows how they relate.
- memory graph
- A way to store information as connected pieces instead of one long block of text.
- retrieval
- The step where a system finds the most relevant text for a question.
- traversal
- The process of moving through connected information to find what is needed.
- AI agents
- AI agents are AI tools that can carry out steps toward a goal, not just answer once.
- AI agent
- An AI program that can inspect information and suggest what to do next.
- lock-in
- A situation where it becomes hard to switch away from one tool or provider.
- traction
- Proof that real people or companies are using or paying for a product.