Step 1: Project Setup

Create the Project

npx create-kernova-app devops-agent
cd devops-agent

Add your API key:

OPENAI_API_KEY=sk-proj-your-key

Create the File Structure

We need more directories than a simple project:

mkdir triggers
mkdir channels
mkdir workflows
mkdir tasks

Your final structure:

devops-agent/
├── agents/
│   └── devops.agent.ts       ← Main agent (schedule, triggers, permissions)
├── agents/tools/
│   ├── check-health.ts       ← Hit /health endpoints
│   ├── get-metrics.ts        ← Read CPU/memory/requests
│   ├── restart-service.ts    ← Restart (requires approval!)
│   └── web-search.ts         ← Default web search
├── triggers/
│   └── on-alert.ts           ← React to infrastructure alerts
├── channels/
│   └── webhook.ts            ← Receive external webhooks
├── workflows/
│   └── incident-response.ts  ← Multi-step durable workflow
├── tasks/
│   └── send-notification.ts  ← Background notification task
├── memory/
│   └── incidents.ts          ← Remember past incidents
├── app/
│   └── chat/page.tsx         ← Dashboard/chat interface
└── kernova.config.ts

Delete the Default Agent

rm agents/assistant.agent.ts

Understanding This Architecture

This project is different from the previous tutorials:

Previous tutorials	This tutorial
Agent waits for user messages	Agent acts on its own (schedule + triggers)
Single interaction model	Continuous monitoring model
No approval needed	Dangerous actions need human approval
Simple request → response	Multi-step durable workflows

The DevOps agent is always running. It checks health every 5 minutes, generates reports at 9am, and responds to alerts instantly — without anyone typing a message.

Key Concept: Autonomous vs Reactive

Most AI agents are reactive — they respond when you ask. This agent is autonomous — it acts on its own based on:

Schedules — “Check health every 5 minutes”
Triggers — “When an alert fires, investigate”
Channels — “When a webhook arrives, process it”

The user still can chat with it (to ask questions or approve actions), but the agent doesn’t need the user to initiate work.

Continue to Step 2: Define the Agent →