#2626th of 27 in Memory
m_flow
Graph-based memory engine that scores retrieval by evidence paths
Documentation ↗
(opens in a new tab)Website ↗
(opens in a new tab)Repository on GitHub ↗
(opens in a new tab)
- Stars
- 4.5k
- License
- Apache-2.0
- Last commit
- Aug 2026
- Last release
- Apr 2026
- Language
- Python
Missed the latest GitHub check; these facts may be older.
Overview
M-flow is a Python memory layer for AI agents that ingests text and files into a four-level graph (Episode, Facet, FacetPoint, Entity). Queries are anchored by vector search at the matching granularity, then scored by graph path cost, and whole Episode bundles are returned. It ships a Python API, a CLI, a FastAPI backend, a Next.js console and an MCP server.
Who it is for: Developers adding long-term graph memory to agents or RAG apps
Strengths
- Multi-granularity graph retrieval; reports 81.8% on LoCoMo-10 and 89% on LongMemEval
- Resolves pronouns to named entities before indexing, so later turns stay retrievable
- Supports LanceDB, Neo4j, pgvector, ChromaDB, KùzuDB and Pinecone backends
- Includes MCP server, CLI, web console and 50+ file format ingestion
Weaknesses
- Benchmark figures are self-reported and compared against vendor cloud products
- Graph building runs through an LLM at ingestion, so it needs an API key or local model
- Summaries keep all details, so retrieved context is longer than with compressed approaches
- Face-aware playground needs a separate companion service and a camera
What it needs
- Docker + Compose
- Needs LLM API key, LanceDB, Neo4j, PostgreSQL/pgvector, ChromaDB, KùzuDB or Pinecone
- Models: OpenAI, Anthropic, Mistral, Groq, Ollama
Also in Memory
See all 27| Rank | Project | Score |
|---|---|---|
| 1 | hindsightAgent memory server with retain, recall and reflect operations | 95 out of 100 |
| 2 | gbrainMemory store with sources and a knowledge graph for coding agents | 91 out of 100 |
| 3 | MemPalaceLocal verbatim memory store for AI agents with semantic search and MCP | 90 out of 100 |
| #4 | SupermemoryMemory and context engine for AI agents, with user profiles and hybrid search | 89 out of 100 |
| #5 | Mem0Long-term memory layer that stores and retrieves user facts for LLM agents | 88 out of 100 |