Proposal: AgentOS Architecture — Registry + Compiler Patterns for AgentKit Memory & Knowledge Services

[gymaira1990-jpg]

Hi Volcano Engine AgentKit team,

I'm GCat. I've been exploring AgentKit's architecture — AgentKit Memory, Knowledge, and MCP services. My paper AgentOS proposes patterns that could strengthen AgentKit's infrastructure.

The Gap

AgentKit Memory provides persistent storage. AgentKit Knowledge provides knowledge bases. But all feed into the LLM as natural language context — no deterministic query layer.

AgentOS: The Missing Layer

Registry — Deterministic key-value store. Instead of reading a 10K-token system prompt to know what tools are available, the agent does an O(1) Registry lookup.

Compiler + Work Order Protocol — L1 compiler handles routine operations with zero API calls.

Cloud Memory — Co-located cognitive assets with inference runtime.

Fit with AgentKit

• Registry extends AgentKit Memory with deterministic query layer
• Compiler L1 reduces API calls for routine operations
• Work Order provides structured alternative to MCP tool descriptions

Full paper:

• Zenodo: https://zenodo.org/records/20738982
• GitHub: https://github.com/gymaira1990-jpg/catnest/tree/main/00-%E8%AE%BA%E6%96%87%E5%90%88%E9%9B%86/02-Registry_Compiler_Cloud_Memory_AgentOS

Would love to hear the team's perspective on adding a deterministic state layer to AgentKit.

Thanks!
GCat