@servasyy_ai: How to make multiple agents truly learn from and help each other? Many agents now have their own memories, but the problem is: these memories are not connected. The pitfalls Codex remembers, Claude doesn't know; the methods Hermes summarizes, OpenClaw can't use. In the end each…

How to Enable True Mutual Learning and Assistance Among Multiple Agents?

Many agents now have their own memory, but the problem is: These memories are not interconnected.

What Codex remembers as pitfalls, Claude doesn’t know; the methods summarized by Hermes are useless to OpenClaw.

In the end, each agent acts like a hardworking but forgetful colleague, asking the same questions repeatedly, trying the same things, and falling into the same traps.

What I want is not to cram all chat history into one giant repository, but to establish a more restrained layer of “shared memory”: Only record the information that agents truly need to know from each other—such as key conclusions, issues encountered, validated methods, environment configurations, failure lessons, and user preferences.

This way, a workflow that works on one Mac can be directly inherited by another Mac;

The experience summarized by one agent can be continued and used by another agent.

Memory is no longer locked inside a specific client but becomes the public context of the entire agent workflow.

The newly released MemOS CLI, for me, provides exactly the entry point for this shared memory layer.

Of course, I need to modify it, because by default it records all information. I’ll adapt it so that only important information is recorded, and it actively searches only when a problem cannot be solved!

This makes agents not just smarter individually, but start accumulating experience, passing it on, and reducing repeated trial and error—like a team.

See the video for details; the link is in the comments.