m_flow
A bio-inspired cognitive memory engine — a new paradigm for Graph RAG.
What is m_flow?
A bio-inspired cognitive memory engine — a new paradigm for Graph RAG.
Overview of m_flow
What it is
M-flow is a bio-inspired cognitive memory engine that introduces a new paradigm for Graph RAG. It organizes knowledge into a four-level cone graph (Episode, Facet, FacetPoint, Entity) and uses graph propagation to score evidence paths, rather than relying solely on similarity matching.
Key features
- Graph-routed bundle search: retrieval is driven by path-cost propagation through a structured memory graph, not by one-shot similarity.
- Four-level cone graph: knowledge is stored at multiple granularities (Episode, Facet, FacetPoint, Entity) connected in a single graph.
- Granularity-aligned anchoring: a query lands on the most precise matching layer, then the graph routes to a coherent Episode bundle.
- Association as controlled propagation: evidence spreads through typed, semantically weighted edges with added cost, ensuring only coherent paths remain competitive.
- Unified multi-granularity retrieval: users do not need to choose a memory layer; the system finds the right anchor and returns the strongest supporting path.
Use cases
M-flow is designed for retrieval tasks where relevance depends on coherent chains of evidence rather than keyword or vector similarity. It is suitable for applications requiring structured memory recall, such as question answering over complex documents, incident analysis, decision process retrieval, and any scenario where a query needs to trace associations across multiple levels of detail.
Getting started
See the repository README for installation and quick-start instructions.
License and pricing
M-flow is open source under the Apache-2.0 license.
Based on the project's public README and repository metadata.