XNews

Memmy Open-Source Agent Memory Management Tool

Memmy Open-Source Agent Memory Management Tool

Memmy, an open-source project, addresses the challenge of unified memory management across different AI agents. It features a three-layer memory system: L1 factual memory, L2 procedural memory, and L3 pattern recognition, which can be reused as skills. Memmy scans and manages conversation histories from agents including Codex, Claude Code, Workbuddy, and Hermes. The tool supports Mac and Windows, and provides both TUI and CLI interfaces. Agents can call and read memories without opening the software, enabling a personal context system. Memmy automatically extracts and creates reusable skills from agent conversations. Users can search all agent conversation histories at any time.

Key insights

Three-Layer Memory Architecture Mirrors Human Cognitive Patterns

Memmy implements a three-tier memory system designed to correspond to human memory modes. L1 handles factual memory, L2 manages procedural memory, and L3 performs pattern recognition leading to reusable skills. The system automatically extracts and creates reusable skills from agent conversations. This architecture allows agents to build upon past interactions rather than starting from scratch each session.

Cross-Agent Compatibility and Multi-Interface Access Lower Adoption Barriers

Memmy supports unified management of conversation histories from multiple agents including Codex, Claude Code, Workbuddy, and Hermes. The tool runs on both Mac and Windows and offers TUI and CLI interfaces. Agents can access stored memories without requiring the Memmy application to be open, enabling integration into existing workflows. Users can search all agent conversation histories at any time, creating a comprehensive personal context system.

Source posts