M-flow
A bio-inspired cognitive memory engine that scores retrieval by the strongest chain of reasoning connecting a query to an answer, instead of ranking by vector similarity or a static knowledge graph.
M-flow is a Graph RAG system where the graph itself acts as the scoring engine rather than a supporting structure: a query first casts a wide net through vector search to find entry points, and the graph then propagates evidence along typed, semantically weighted edges, scoring each knowledge unit by the strongest chain of reasoning that connects it back to the query.
It distinguishes similarity, proximity in representation space, from relevance, whether the system can trace a coherent chain of evidence from query to answer, and stores knowledge in a four-layer cone graph of episode, facet, facet-point and entity, so a query lands on whichever layer matches its granularity. An accompanying OpenClaw skill lets the engine be used from that agent directly.