Hmem v2: Persistent Hierarchical Memory for AI Agents

Hmem v2 addresses the problem of AI agents forgetting information between sessions by providing them with persistent, hierarchical memory. As a Model Context Protocol (MCP) server, Hmem enables AI agents to retain knowledge across multiple machines or sessions, preventing them from duplicating work and contradicting previous decisions.

Unlike existing RAG solutions that offer flat memory structures, Hmem stores memory in five nested levels of detail, mirroring how human memory works. This hierarchical structure allows agents to load only the context they need, avoiding token waste and optimizing context retrieval. A freshly spawned agent receives only a coarse summary (Level 1) and can request more detail as needed.

Hmem also includes features for updating and appending memory entries, as well as marking entries as obsolete. A dedicated curator agent runs periodically to maintain memory health, detecting duplicates, merging fragmented entries, and pruning low-value content. This comprehensive memory management system aims to provide AI agents with a more human-like memory experience, enhancing their learning and reasoning abilities.

Transparency Disclosure: This analysis was prepared by an AI language model, Gemini 2.5 Flash, based on the provided source content. The AI model has been trained to provide objective summaries and insights, but its analysis should be considered as one perspective among many. The user retains full responsibility for decisions made based on this information. This content is compliant with EU AI Act Article 50 regarding transparency.