memU
NevaMind-AI/memU
Shared LLM wiki memory across sessions, agents and devices that also distills reusable skills
memU is a lightweight, agent-driven memory system that gives users a shared LLM wiki across sessions, agents and devices. Its tested-integration matrix lists Hermes Agent as supported on macOS and Windows.
What memU does
memU gives users a shared LLM wiki across sessions, agents and devices and automatically distills reusable skills from agent history. The README says its core memory logic is only 500 lines, compact enough to inspect and adapt. A scheduled background task captures new session history, including messages and tool calls, slices it into self-contained jobs and lets the agent decide whether to do nothing, patch an existing skill or create a new one as readable Markdown.
Setup is agent-driven: you get an API key from memu.so and send your agent a message telling it to read https://memu.pro/SKILL.md and follow the instructions. The README's matrix lists Hermes Agent as supported for both memorize and retrieve on macOS. On Windows memorize works and retrieve carries a warning that a memU version with Windows HERMES_HOME support is needed. Hermes Agent does not appear in the Linux table. Other listed hosts include Codex, Claude Code, Cursor, OpenClaw and WorkBuddy.
Key features
- Shared LLM wiki across sessions, agents and devices
- Automatic extraction of reusable Markdown skills from agent history
- Scheduled background memorize task and retrieve into future tasks
- Per-OS support matrix covering macOS, Windows and Linux hosts
- Agent-driven install from a SKILL.md instruction file
- Core memory logic of about 500 lines
When to use it
- Letting Hermes Agent carry knowledge between sessions and machines
- Turning repeated workflows in agent history into reusable skills
- Sharing one memory between Hermes Agent, Claude Code and Cursor
Who it is for: People who use several coding agents and want one shared, inspectable memory that also builds skills.
How it fits with Hermes Agent
Hermes Agent is listed as a tested host: memorize and retrieve work on macOS, Windows retrieve needs a memU version with HERMES_HOME support, and Hermes is not in the Linux table.
Requirements: An API key from memu.so and a supported host agent
Note: On Windows, Hermes Agent retrieval needs a memU version with Windows HERMES_HOME support, and Hermes is not listed in the Linux support table.
FAQ
What is memU?
memU is a lightweight, agent-driven memory system that keeps a shared LLM wiki across sessions, agents and devices. It also distills reusable skills automatically from your agent history.
Does memU work with Hermes Agent?
Yes, with platform limits. The README lists Hermes Agent as supported for memorize and retrieve on macOS. On Windows retrieve needs a memU version with HERMES_HOME support, and Hermes is not listed for Linux.
What do I need to run memU?
You need an API key from memu.so and a supported host agent such as Hermes Agent, Claude Code, Codex or Cursor. You then send your agent a message to read https://memu.pro/SKILL.md and follow its instructions.
Similar memory for Hermes Agent
All memoryMemory and context engine for AI agents with an MCP server, open-source plugins and a Hermes provider
garrytan GBrainAgent memory that stores sourced facts and shares them across agents over MCP, built by Garry Tan
TencentCloud TencentDB Agent MemoryTeam-level memory hub that shares chat memory, skills, wiki and code graph across agents through a proxy
screenpipe screenpipeRecords screen and audio locally and exposes searchable computer history to agents via CLI skills and MCP
MemoriLabs MemoriLLM-agnostic memory layer with Python and TypeScript SDKs that persists and recalls agent conversations
MemTensor MemOSMemory operating system for LLMs and agents with official local plugins for Hermes Agent and OpenClaw
Related guides: SOUL.md for Hermes Agent: what it is and how to write one · Run multiple Hermes agents with profiles