m_flow

m_flow

A bio-inspired cognitive memory engine that performs reasoning and associative retrieval through graph structures.

MCPDataOpen sourceGraphRAG认知记忆推理检索
Type
MCP
Transport
http
Open source
Yes
GitHub Stars
★ 4.4k
Source
mcp-github

Overview

M-flow is a bio-inspired cognitive memory engine that conducts reasoning and associative retrieval via graph structures. The system organizes knowledge into a four-layer conical graph, ranging from abstract summaries to atomic facts. When a query arrives, vector search broadly identifies entry points across multiple granularities, after which the graph structure takes over—propagating evidence along typed, semantically weighted edges and scoring each knowledge unit based on the strongest inference chain. This mechanism enables M-flow to locate relevant answers more accurately than simple keyword matching, making it suitable for applications requiring complex contextual understanding and reasoning.

Capabilities

  • Multi-granularity vector search
  • Graph-based reasoning
  • Semantic-weighted edge propagation
  • Strongest inference chain scoring
  • Four-layer conical graph organization of knowledge

Use cases

Complex context understandingKnowledge graph retrievalEvent causality analysisDecision process review

Setup

Requires: Python 3.10-3.13API Key
pip install m_flow

This information was compiled by AI from public sources and may contain inaccuracies — please refer to the source.

FAQ

How is M-flow different from traditional RAG systems?

M-flow performs reasoning and associative retrieval through graph structures, rather than relying solely on keyword matching.

What is the four-layer conical graph in M-flow?

The four-layer conical graph consists of Episode, Facet, FacetPoint, and Entity, representing events, dimensions, atomic assertions, and entities, respectively.

Related skills