FlowElement-xinliuyuansu/m_flow

Health v2 · maintenance only

Flags: young

Adoption not part of the score

4500 stars · 257 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

M-flow is a Python bio-inspired cognitive memory engine that treats graph topology as the retrieval scoring engine rather than a supportive structure. It stores knowledge in a four-layer cone graph (Episode → Facet → FacetPoint → Entity) and scores relevance by propagating evidence along typed, semantically weighted edges from vector-search entry points.

Use cases

• build long-term memory for ai agents
• implement graph rag with reasoning-based retrieval
• give llm chatbots episodic and semantic memory
• retrieve answers by evidence paths instead of vector similarity
• add multi-granularity memory recall to rag pipelines
• expose agent memory via mcp server

When to choose

• you need relevance based on reasoning chains rather than embedding similarity
• you're building agentic ai that must remember and associate events over time
• you want a graph-native alternative to chunk-based GraphRAG systems
• you need MCP-compatible memory for LLM tooling

When to avoid

• you need a simple, battle-tested vector-only RAG stack
• your corpus is small enough that plain similarity search suffices
• you require a managed database product rather than a Python library
• you're outside the Python 3.10–3.13 ecosystem

Facets

library · maturity active

rag vector-database search-engine agent-framework mcp llm-inference artificial-intelligence large-language-models databases python cross-platform graph-rag agent-memory knowledge-graph episodic-memory semantic-memory memory-engine cognitive-architecture evidence-path-scoring retrieval-augmented-generation ai-agents natural-language-processing

Member repositories

For agents

markdown · JSON · MCP: product_card(name="FlowElement-xinliuyuansu/m_flow")