跳到主要内容

Architecture

NPO Studio is a core component of the SK NPO platform, designed to facilitate the creation, management, and deployment of AI agents and models. It sits between NPO Service (AI Ops/Workspace) and the underlying NPO Core infrastructure, providing tools and environments to build and manage AI-driven business solutions.

Four Key Elements​

1. AIOps, AI Workspace (NPO Service)​

The top-level service layer providing an AI-powered work processing environment.

  • AI Workspace (ChatOps): Natural language-based business processing through chat.
  • Connects three core components: Agents, System Tools, and Knowledge.
  • Supports Service AI integrations for end-to-end business workflows.

2. NPO Studio (Agent/MCP Builder)​

The builder layer for creating agents and integration tools.

  • Agent Builder: Design and configure AI agents with visual tools.
  • Agent Catalog: Browse and discover registered agents.
  • MCP Builder (API-based): Create MCP Server profiles to connect agents with backend APIs.
  • MCP Registry: Manage and discover MCP server registrations.
  • RAGaaS: Retrieval-Augmented Generation as a Service for knowledge (vector) management.
  • External Agent/MCP: Connect third-party agents and MCP servers.

3. Model Manager​

Management and serving layer for language models.

  • Supports Open LLM, Pre-trained, and Model Serving workflows.
  • Playground for model testing.
  • Targets small/medium-sized models for 100 to 1,000 users (e.g., LLama 3.1 8B, Mistral 7B, Stable Diffusion, Whisper Large, YOLOv8m).

4. GPUaaS​

GPU resource management layer.

  • GPU Pool and L/C Management (Lifecycle Management).
  • GPU profile and pool management and scheduling.
  • GPU allocation, monitoring, and billing.

NPO Core (Infrastructure)​

The foundation layer providing core platform services:

ComponentCapabilities
AI DevSecOps (AMDP)Build/Deployment Dashboard, CI/CD Pipeline (Template, Configuration)
API Management (APIM)API Security (WAF), API Gateway & API Management
Database (ZDB)DB Provisioning (6 types), DB Lifecycle Management
Container Platform (ZCP)Multi-Cluster Management, Kubernetes Engine

The entire platform runs on underlying compute and cloud infrastructure (bare-metal servers and cloud environments).