CodeAbra/iai-personal-memory-engine

What it does

CodeAbra's Personal Memory Engine is a local memory daemon designed for Model Context Protocol (MCP) clients. It enables persistent, encrypted personal memory storage with a three-tier architecture: episodic memory (experiences), semantic memory (knowledge), and procedural memory (skills/habits). The tool integrates vector search capabilities using hnswlib with BGE-small-en-1.5 embeddings for intelligent memory retrieval and ranking.

How to set up

This is a Python-based MCP server that runs locally. Install from the GitHub repository, configure your MCP client to connect via stdio, and initialize the SQLite database. The tool uses AES-256-GCM encryption for secure local storage. Configure your preferred embedding model and vector database parameters before first use.

Key features

• Three-tier memory architecture (episodic, semantic, procedural)
• Local SQLite database with hnswlib vector store
• AES-256-GCM encryption for data security
• BGE-small-en-1.5 embeddings for semantic search
• MOSAIC community detection reranking
• Sleep cycle consolidation for memory organization
• Cross-platform support (Windows, macOS, Linux)