MemU

A layer of personal memory shared by multiple AI agents: Codex, Claude Code, Cursor, etc. share the same readable Markdown memory file, supporting source tracing, memory retrieval, and automatic skill extraction.

AI Agents#AI Agent#Task Automation#Universal Assistant

Updated: 2026-10-26

What is MemU?

MemU solves a very specific problem: every time an Agent starts a new session, it loses its memory. It connects clients such as Codex, Claude Code, Cursor, OpenClaw, Hermes, and WorkBuddy to the same layer of personal memory, enabling cross-session, cross-agent, and cross-device memory sharing. Moreover, the memory is not stored in a black-box vector database but in directly readable Markdown files. The official name for this design is the "Inspectable Memory Layer."

Who is it for?

Developers who use multiple coding Agents at the same time: They don't want to repeat the project constraints in each tool.
Independent developers who maintain the same project long-term: They need to remember "why the decisions were made in the first place."
Small teams: Settle project preferences and decision records into shared documents instead of scattering them across chat logs.
Freelancers managing multiple client projects simultaneously: Create memory files per project to avoid context mix-ups when switching between them.
People who care about data ownership: They want their memories to be stored as plaintext files in their own directory.

Key features

Cross-Agent Sharing

The official states that one memory layer is available for all Agents to use, eliminating the need to re-explain the project background when switching tools.

Traceable Source

Each memory entry comes with Source, Updated, and Related fields, allowing you to review which conversation it originated from and when it was written.

Readable Markdown

Memories are presented as files like project-preferences.md and launch-decisions.md, which can be directly modified when issues arise.

Automatic Skill Extraction

Extract habits and tool usage from historical tasks, and summarize them into "pre-task" and "post-task" operation points as well as "things to avoid".

Memory Retrieval and Storage

Retrieval and storage are divided into two layers, which are used on demand instead of stuffing all context into prompts.

Multi-device Sync

The same memory is shared across laptops, office desktops, and home servers, so there's no need to re-enter project background when switching devices; the official provides the installation method for SKILL.md, which uses an API Key as a Bearer token to pull instructions and complete configuration.

How to use MemU

Only sync the "Project Preferences" and "Decision Records" categories first, and don’t include temporary debugging information—this will greatly improve retrieval accuracy.
Regularly clean up expired entries: Markdown has the advantage of being readable, but the downside is that it’s easy to accumulate a mess.
An API Key is like the key to the memory.
do not commit it to public repositories.
assign it to individuals on a per-need basis and revoke it as required during collaborative work.
The memory may contain customer information and internal decisions, so please confirm its data usage terms before connecting to third-party Agents.
When a team shares a single memory directory, first agree on who can modify which files to prevent overwriting each other.
when setting up on a new device, install and verify the retrieval results on one machine first before deploying to other devices, so as to avoid copying incorrect configurations everywhere.
The official says it can be freely copied and pasted into Agents for 100% current use, and the paid usage and quota rules are subject to the in-site instructions.