What it does
Memory is organised in five levels. Level 1 holds one-line summaries that load at session start, and deeper levels hold the complete original text, loaded on demand. Conversations are recorded automatically. Every N exchanges (default 5), a Haiku subagent runs in the background, extracts lessons, errors and decisions, and writes a handoff note to the project. Memory is tied to projects rather than sessions. The package also includes 21 skill files that agents load as needed, covering session start, reading, writing, search, curation and migration.
It suits projects where an agent should recall past mistakes, decisions and open tasks when work resumes, including on another device or with another AI provider. The publisher describes it as usable with Claude, Gemini, GPT, local models and any MCP client. Loading a project briefing is described as costing about 3,000 tokens.
Setup is documented for Claude Code, Gemini CLI and OpenCode, and updates run through npm update -g. Cross-device sync is configured separately with hmem-sync. Settings, hooks and checkpoints are managed through a configuration skill.
Limitations and requirements
Cross-device sync needs a separate hmem-sync setup. First-time installation is documented for Claude Code, Gemini CLI and OpenCode. Updates are performed with npm.
Safety
Our team has not reviewed this resource. Only automated rules checked it; resources with high-risk signals are not published. These are signals, not a guarantee that installing it is safe. Try it in a sandbox project first.
- Runs code or changes an assistant's behaviour
Sources
- Official MCP Registry first seen 13 Nov 2026 · last seen 13 Nov 2026 · its page on the source
hmem — Humanlike Memory for AI Agents
Persistent 5-level hierarchical memory for AI agents. SQLite-backed, lazy-loaded.
v1.6.7As shown on registry.modelcontextprotocol.io when we last read it