microsoft / microsoft/agent-framework
Proposal: AgentOS — An OS-Inspired Architecture for Agent Memory, Communication & State Management
@sphenry is already working on this.
Since Jun 18, 2026.
- Dominant language
- Python
- Stars
- 13.6k
- Forks
- 2.3k
- Avg merge
- 2d 45m
- Merged PRs (30d)
- 358
Description
Hi Agent Framework team,
I'm GCat. I recently published a technical white paper proposing **AgentOS** — a complete agent operating system architecture.
## The Problem
Today's agent frameworks treat agent memory as "context injection" — system instructions, conversation history, RAG snippets, and tool outputs are all mixed into a linear context window.
## AgentOS: Three Core Pillars
**1. Four-Table Local Memory** — Registry (O(1) deterministic config), Environment Variables (runtime), System Log (audit), Memory Store (unstructured data).
**2. Compiler + Work Order Protocol** — L1 rule-based (zero token cost) + L2 model-based compiler translates intent into structured JSON Work Orders.
**3. Cloud Memory Pointer Service** — Cognitive assets co-located with inference engines via cache-handshake.
## Why Agent Framework
- Registry as deterministic config layer
- Work Order Protocol for inter-agent communication (A2A complement)
- Cloud Memory for persistent context
Full paper (V3.1, complete specs):
- 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
Thanks!
GCat
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Assessment
This issue has not been assessed yet.