When to use it#

llama-index-agensgraph is for teams building retrieval and RAG applications in LlamaIndex that want AgensGraph behind LlamaIndex’s standard index abstractions.

Good fits

  • Property-graph RAG. Build a PropertyGraphIndex from documents with SchemaLLMPathExtractor, then answer questions with combined graph + vector retrieval, or with AgensGraph-dialect Text2Cypher, run read-only by SafeTextToCypherRetriever.

  • Vector RAG. Back a VectorStoreIndex with pgvector and query it through any LlamaIndex query engine or retriever.

  • Precise, filtered retrieval. Constrain results with LlamaIndex MetadataFilters (all 14 operators, plus AND/OR/NOT and nested groups), index the keys you filter on, and combine semantic and keyword search with reciprocal-rank-fusion hybrid mode, filtered or not.

  • Mixed-retrieval apps. Route a single question between a graph index and a vector index, sharing one AgensEngine pool, with a RouterQueryEngine or an agent.

  • Async services. Use the async methods (aquery, async_add, aupsert_nodes, avector_query, …), which do not block the event loop, for concurrent workloads.

How it compares

  • Pick llama-index-agensgraph when your application is built on LlamaIndex.

  • Pick LangChain if you build on LangChain instead — it also adds chat history and a LangGraph checkpointer.

  • Pick LightRAG or cognee for a turnkey graph-RAG or memory pipeline rather than assembling indices and retrievers yourself.

Consider an alternative when

  • You want an agent such as Claude Desktop to use AgensGraph through tools with no application code — use the MCP servers.

  • You only need raw Cypher access without LLM features — the agensgraph-python driver is enough.