Hmem — Humanlike Memory For AI Agents
Give Claude Code or Cursor a SQLite-backed memory layer so long-running agent work survives restarts without re-pasting context.
Overview
hmem is a Build-phase MCP server that provides persistent five-level hierarchical SQLite memory for AI agents.
What is this MCP server?
- Five-level hierarchical memory model tuned for humanlike recall patterns in agents
- SQLite persistence with lazy loading to keep token use lower on large histories
- stdio MCP package via npx (hmem-mcp v1.6.7) with configurable HMEM_STORAGE_DIR
- Local default storage at ~/.hmem/ for offline, builder-controlled data
- Persistent context across sessions without custom vector DB or hosted memory SaaS
- 5-level hierarchical memory model
- Server version 1.6.7 (npm hmem-mcp)
- Default storage directory ~/.hmem/
Community signal: 20 GitHub stars.
What problem does it solve?
Agent chats forget prior decisions and project facts every session, forcing solo builders to repeat context or build bespoke note pipelines.
Who is it for?
Solo builders running Claude Code or Cursor on multi-session repos who want local, hierarchical agent memory without a hosted memory product.
Skip if: Teams that need shared team-wide memory in the cloud, strict compliance-hosted vector search, or one-off scripts with no recurring agent use.
What do I get? / Deliverables
After registration, your agent can read and write structured long-term memory locally so follow-up tasks reuse prior conclusions without manual paste-ins.
- Registered stdio MCP server exposing hierarchical memory tools
- On-disk .hmem SQLite databases under default or custom storage dir
- Reusable agent recall across coding sessions on the same machine
Recommended MCP Servers
Journey fit
Agent memory is wired during product and tooling setup in Build, then pays off across every later phase of the solo-builder journey. Canonical shelf is agent-tooling because hmem is installed as an MCP stdio server alongside other agent integrations, not as app UI or backend feature code.
How it compares
MCP memory integration, not a brainstorming or planning agent skill.
Common Questions / FAQ
Who is hmem for?
Indie and solo developers who use MCP-enabled coding agents and need durable, local memory across sessions.
When should I use hmem?
Use it when you are building or operating agent workflows where architecture choices, user prefs, or task state must survive chat resets.
How do I add hmem to my agent?
Add the hmem-mcp stdio server to your MCP client config (npx hmem-mcp), optionally set HMEM_STORAGE_DIR, then restart the agent so memory tools are available.