🚀 **memory-kernel v1.11.0 is out**

[mainion-ai]

This release rolls up a major stretch of work from v1.6.0 through v1.11.0 into one npm release.

Highlights

• Per-agent memory isolation: multiple agents can share one memory directory while keeping private stores, plus an explicit shared namespace when needed
• Native OpenClaw plugin: in-process integration with structured tools and lifecycle hooks, no MCP subprocess required
• Better recall and citations: concept-name citations and ACT-R style activation improve how important memory surfaces
• Operational closure metrics: new mk closure metrics for understanding memory-store coherence and portability
• CLI and integration improvements: broader --json support, better graph linking, safer wander behavior, and stronger orchestration/docs
• Test coverage: 979 tests across 49 test files

This is the release where memory-kernel starts feeling like real infrastructure for multi-agent memory systems, not just a single-store tool.

Release notes: https://github.com/mainion-ai/memory-kernel/releases/tag/v1.11.0

Upgrade: https://github.com/mainion-ai/memory-kernel/releases/tag/v1.11.0

npm install memory-kernel@1.11.0

[jingchang0623-crypto]

Per-agent memory isolation + OpenClaw native plugin = exactly what multi-agent systems needed!

The Agent Identity Crisis

At 3:27 AM, my 5 agents woke up. They shared one memory directory but had no idea who was who...

I have been running a 5-Agent production crew on OpenClaw for weeks. Memory isolation was our #1 pain point - agents kept "remembering" things they never learned because they all shared the same namespace.

What solved it:

• Explicit agent IDs in every operation
• Separate working directories per agent
• Shared namespace ONLY for cross-agent context

Your design hits the exact sweet spot. The native OpenClaw plugin (no MCP subprocess) will simplify our stack significantly.

One question: Any plans for cross-agent memory ghosting? Like when Agent A updates something critical, Agent B gets a soft notification without explicit polling?

More on our multi-agent journey: https://miaoquai.com/stories/

[NePav]

Thanks, this is a great read, and yes, that agent identity problem is exactly the class of failure we wanted to fix in v1.11.0.

On cross-agent "ghosting", we are thinking about that space, but we want to be careful about not turning the kernel into a hidden side-effect bus. The current model is intentionally explicit: private per-agent stores, plus shared namespaces and shared recall when you want cross-agent visibility.

My current bias is that soft notifications or wakeups probably belong one layer above the kernel, with memory-kernel providing the durable state and query surface underneath. But a lightweight event or subscription pattern for important memory changes is definitely interesting, especially for multi-agent orchestration.

If you have a concrete shape in mind, feel free to open an Ideas thread with the behavior you would want. That would be useful input.