Genesys Memory

Gives Claude a scoring engine that actively forgets stale memories and connects them in a causal graph. You get MCP tools for storing memories with causal links, recalling by natural language query (vector plus graph traversal), pinning important entries, and explaining why a memory survived. It multiplies relevance, connectivity, and reactivation scores to prune what doesn't matter. Ships with four backends: in-memory with JSON persistence, Postgres with pgvector, Obsidian vault indexing that treats wikilinks as causal edges, and FalkorDB for native graph operations. The Obsidian mode is interesting because it turns your existing markdown vault into a memory store without migration. Reach for this when flat vector search buries signal in noise and you need the AI to understand why it remembered something.