PlateerLab/synaptic-memory

Distributes graph structure to Neo4j, vectors to Qdrant, and content blobs to MinIO for production multi-agent deployments. Built as a pure-Python library with modular backends (in-memory, SQLite, Neo4j + Qdrant), it auto-constructs ontologies using rule-based classification, semantic embeddings, and optional LLM enrichment, then ranks retrieval across five dimensions (relevance, importance, recency, vitality, context). Exposes 16 tools via MCP server and implements four-tier memory consolidation (L0–L3) with Hebbian co-activation weighting to reinforce successful agent decision patterns.

Available on PyPI.