CodeAbra/iai-mcp

What it does

CodeAbra/iai-mcp is a local memory daemon designed to augment AI systems with persistent, encrypted knowledge storage. It implements a sophisticated three-tier memory architecture—episodic (event-based), semantic (conceptual), and procedural (skill-based)—allowing AI tools like Claude to maintain and recall contextual information across sessions.

Key features

• Three-tier memory storage: episodic, semantic, and procedural
• Vector search powered by LanceDB with BGE-Small-EN-v1.5 embeddings
• Graph-based recall reranking for improved relevance
• Sleep-cycle consolidation for memory optimization
• AES-256 encryption at rest for data security
• Verbatim recall accuracy ≥99% at 10k records
• Sub-100ms p95 latency for responsive performance
• Ambient capture integration via Claude Code Stop

How to set up

Installation requires Python 3.8+ and involves cloning the repository, installing dependencies via pip, configuring the daemon for your platform (Linux, macOS, or Windows), and integrating with your MCP-compatible AI client like Claude.