Hmem

About

AI agents often lose all context and progress when a session ends or work shifts across different environments. Hmem addresses this by providing agents with persistent, humanlike memory, structured into five lazy-loaded levels of detail. This innovative approach allows agents to retrieve exactly the amount of context needed—from broad summaries to fine-grained specifics—ensuring continuity, eliminating duplicate effort, and preserving hard-won insights across all sessions and machines.

Key Features

• Hierarchical memory retrieval with lazy loading across five levels of detail
• 0 GitHub stars
• True tree structure for memory, allowing multiple siblings at the same depth
• Persistent memory across sessions, ensuring agents retain context after restarts
• Per-agent memory stored in individual SQLite files, with an option for shared company knowledge
• Native Model Context Protocol (MCP) compatibility, working with tools like Claude Code and OpenCode

Use Cases

• Maintaining continuous context for AI agents across multiple development sessions or machines
• Enabling multi-agent AI systems to share and build upon persistent knowledge
• Optimizing token usage by providing AI agents with context-aware, lazy-loaded memory on demand