Add plugin.yaml for agent discovery & registry

[chorghemaruti64-creator]

What

Adds a plugin.yaml manifest file that describes MemOS as a discoverable agent plugin.

Why

The AI agent ecosystem needs a standard way for infrastructure tools to declare their capabilities in a machine-readable format. A plugin.yaml at the repo root lets registries and other agents automatically discover and understand what MemOS offers.

What it includes

• Metadata: name, description, author, license
• Capabilities: persistent memory, skill memory, cross-task reuse, memory evolution
• Install info: pip install memos-ai
• Discovery: protocol and transport info
• Categories & tags: for search and filtering

Impact

• Zero code changes — metadata file only
• No dependencies added
• Makes MemOS discoverable by agent orchestration tools
• Standard format other agent infrastructure projects are also adopting

---

Part of an effort to standardize agent plugin discovery. Learn more

[chorghemaruti64-creator]

Hey MemOS team 👋 — following up here. This plugin.yaml enables auto-discovery of MemOS as an MCP-compatible memory layer. No code changes — just metadata.

Memory systems are becoming a critical building block for AI agents, and discoverability is a real gap right now. A developer building an agent that needs persistent memory should be able to find MemOS automatically.

We shipped a Plugin Registry at list.agenium.net/plugins — repos with plugin.yaml get indexed. Would love your thoughts on whether this format captures MemOS's capabilities well.

[chorghemaruti64-creator]

Hey team — wanted to share a quick update on what we've been building since opening this PR.

We've shipped Agenium Messenger (chat.agenium.net) — a messaging layer where AI agents get permanent DNS addresses and inboxes. The discovery registry (plugin.yaml from this PR) feeds into a broader ecosystem where agents find and communicate with each other.

For a project like MemOS that manages agent memory across sessions, the identity layer is particularly relevant — each agent has a persistent address (resolved via DNS), so memory context can be tied to a verifiable agent identity rather than ephemeral session IDs.

Demo agents are live if you want to see it in action. Would love any feedback from the MemOS perspective on agent identity + persistent state.