What is this?
memU is a lightweight, agent-driven memory system that gives coding agents a shared LLM wiki across sessions, agents, and devices, automatically turning your agent history into reusable skills.
memU is a lightweight, agent-driven memory system that gives coding agents a shared LLM wiki across sessions, agents, and devices, automatically turning your agent history into reusable skills.Is this a standalone app or does it plug into my existing agent?It runs as a sidecar to a desktop coding agent, with one dedicated binary per host (Codex, Claude Code, Cursor, etc.), rather than being a standalone application.
It runs as a sidecar to a desktop coding agent, with one dedicated binary per host (Codex, Claude Code, Cursor, etc.), rather than being a standalone application.How do I get it running?Grab an API key from memu.so, then send a one-line message to your agent telling it to read memU's docs and follow the instructions to install and configure it; a separate self-hosted flow exists if you want local storage instead.
Grab an API key from memu.so, then send a one-line message to your agent telling it to read memU's docs and follow the instructions to install and configure it; a separate self-hosted flow exists if you want local storage instead.Which coding agents does it support?It has tested adapters for Codex, Claude Code, Cursor, OpenClaw, Hermes Agent, WorkBuddy, Cola, and pi, plus a generic detect mode for other agents.
It has tested adapters for Codex, Claude Code, Cursor, OpenClaw, Hermes Agent, WorkBuddy, Cola, and pi, plus a generic detect mode for other agents.Does memory/retrieval work the same on every agent and OS?No, support varies by host and OS: ChatGPT's Chat mode is unsupported (Work mode only), Claude/Cowork chat isn't supported at all, and some combos like Hermes Agent on Windows or WorkBuddy under Hy3 have known retrieval caveats.
No, support varies by host and OS: ChatGPT's Chat mode is unsupported (Work mode only), Claude/Cowork chat isn't supported at all, and some combos like Hermes Agent on Windows or WorkBuddy under Hy3 have known retrieval caveats.Can I keep my memory fully private and local instead of using memU Cloud?Yes, the self-hosted mode stores memory locally on a single device using your own embedding provider key, instead of syncing through memU Cloud.
Yes, the self-hosted mode stores memory locally on a single device using your own embedding provider key, instead of syncing through memU Cloud.What happens to my memory if I uninstall memU?Uninstalling removes the host integration and tooling but keeps your memory store and config file by default, so a later reinstall can resume where you left off; data is only erased if you explicitly ask for it.
Uninstalling removes the host integration and tooling but keeps your memory store and config file by default, so a later reinstall can resume where you left off; data is only erased if you explicitly ask for it.How does the automatic skill extraction actually work?A scheduled background task slices new session logs into self-contained jobs, the agent itself decides whether to do nothing, patch, or create a skill, and memU embeds and stores the resulting Markdown skill for retrieval on similar future tasks.
A scheduled background task slices new session logs into self-contained jobs, the agent itself decides whether to do nothing, patch, or create a skill, and memU embeds and stores the resulting Markdown skill for retrieval on similar future tasks.Where does my memory data get stored?By default it lives in a local SQLite file at ~/.memu/memu.sqlite3, but you can point it at a Postgres database (with pgvector) for concurrent access or larger stores.
By default it lives in a local SQLite file at ~/.memu/memu.sqlite3, but you can point it at a Postgres database (with pgvector) for concurrent access or larger stores.How complex is the core memory engine?The author kept memU's core memory logic to about 500 lines of code, small enough to read, understand, and modify yourself.
The author kept memU's core memory logic to about 500 lines of code, small enough to read, understand, and modify yourself.What are similar projects?Raven, opensquilla, EverOS, agentmemory, MemOS, QwenPaw, CowAgent, OpenSpace
Raven, opensquilla, EverOS, agentmemory, MemOS, QwenPaw, CowAgent, OpenSpaceWhich topics is it in?loop-engineering, openclaw-skills, embeddings, openclaw, harness, claude-skills, agent-memory, memory, sandbox, skills, sqlite, cloud
loop-engineering, openclaw-skills, embeddings, openclaw, harness, claude-skills, agent-memory, memory, sandbox, skills, sqlite, cloud