Mnemo Mcp

A production-grade persistent memory layer that solves the context problem for long-running AI sessions. Stores facts, preferences, decisions, and skills in SQLite with FTS5 full-text search plus sqlite-vec semantic retrieval, fused via reciprocal rank fusion and reranked by a local Qwen3 cross-encoder. The capture action accepts six typed context categories and deduplicates via embeddings. Includes a temporal knowledge graph with entity resolution, LLM-driven compression of old memories, and encrypted cross-machine sync to Google Drive or S3. Ships with skill definitions that teach Claude when to commit memories and how to recall context proactively. Reach for this when you need the AI to remember conversations across sessions without manually managing context windows or paying for third-party memory APIs.