2026Live when my Mac is on
GraphMind
Turn anything into a knowledge graph you can ask questions of.
A knowledge-graph platform I built end to end: documents, code, audio, video and whole repositories become queryable graphs, with GraphRAG chat and an MCP server for other agents. In a controlled evaluation it answered more faithfully than vector RAG on every model tested — and its graph-first retrieval cut tokens by 59% without losing faithfulness.
GraphRAGNeo4jFastAPICeleryMCPLit
- models where GraphRAG beat vector RAG on faithfulness and precision
- 3 / 3models where GraphRAG beat vector RAG on faithfulness and precision
- multi-hop faithfulness, vs 0.47 for vector RAG (deepseek-r1)
- 0.86multi-hop faithfulness, vs 0.47 for vector RAG (deepseek-r1)
- tokens per answer with graph-first retrieval
- −59%tokens per answer with graph-first retrieval
- faithfulness kept after the token cut (0.94 before)
- 0.93faithfulness kept after the token cut (0.94 before)