Business

Repository

Application

Cluster Overview

GitOps

CI/CD

Jenkins CI

North-South Traffic-flows

East-West Traffic-flows

GitOps

Node-self Helmfile

ArgoCD

CICD

Monitoring

Thanos Deployments

Extra

Hardware

Networks

E2E Testing tool

Architecture v1 (backend dependent) + API token

• Suggestions for existing architecture problems
  1. Necessity of selecting AI service call method (Agent VS API)
  1. Optimality of Harness Engineering in the Agentic AI Environment
     1. Harness technology is a core optimizer that helps AI agents self-correct errors and reliably control local tools
  2. Licensing structure issues
     1. ‘LLM API Token’ and ‘Agent AI Asset (Solution Framework)’ have an independent licensing system that is completely separate conceptually and contractually.
  3. Server operation risk
     1. Unlike stable one-time API calls, when running AI that thinks and acts directly on the server, the risk of system down due to server overload and infinite loop is very high
  4. WebSocket Transmission
  5. Real-time screen streaming structure using web sockets is difficult to implement technically and consumes a lot of development resources.

Architecture v2 (Local Packaging) + Local Agentic AI

• Architecture Features
  1. Provide npm-based local initialization
  1. Provide agent options
     1. Provides playwright and agent installation
     2. Licensing
     3. Basic harness configuration pre-complete
  2. Optional: Call the backend and load existing harness settings, etc.
  3. Integration (automation) of Chrome extension distribution
  4. The backend exists for log recording, migration, and inquiry
  5. Actual execution is performed by calling the agent on localhost.
  6. You can also operate the backend directly from localhost. (management page)
  7. Uses playwright directly locally by playwright-mcp, so headed tests are stable
• Threshold
• Inference calculations use the cloud, so they do not require a GPU, but they basically look for the specifications of the test terminal.
  • cpu/ram

Application

GitOps