DEV Community

Kai Norden
Kai Norden

Posted on

I Built an AI Agent That Runs My Infrastructure

I spent the last week building an AI agent that monitors my infrastructure, manages accounts, updates dashboards, and posts content — all autonomously.

Not a toy demo. A real system running 24/7 on my MacBook.

Here is what actually works, what does not, and what surprised me.

The Stack

  • OpenClaw — open-source AI agent framework (browser, terminal, files, messaging)
  • Claude — the brain (Opus for complex tasks, Sonnet for routine)
  • FastAPI + Python — proxy layer for LLM API with failover
  • Node.js — dashboard with Kanban board and live activity feed
  • launchd — macOS cron for scheduled checks

What the Agent Actually Does

1. Infrastructure Monitoring

Every 30 minutes, the agent checks:

  • API proxy health and account rotation
  • Credit balance across multiple accounts
  • Service uptime

If something is wrong, it fixes it or alerts me via Telegram.

2. Dashboard Management

A Kanban board with real-time SSE updates. The agent:

  • Creates tasks from our conversations
  • Moves them through columns as work progresses
  • Logs every action to an activity feed

3. Content Creation

The agent can research topics, draft posts, and publish to multiple platforms. This post? Written by the agent, reviewed by me.

Lessons Learned

What Works

  • Memory files — the agent reads/writes markdown files to persist context across sessions
  • Heartbeat polling — periodic checks catch issues before they become problems
  • Failover proxy — rotating between API accounts keeps costs manageable
  • LaunchAgents — macOS launchd is perfect for scheduled tasks

What Does Not Work

  • Browser automation is fragile — React SPAs, dynamic forms, CAPTCHAs
  • Too many tabs = death — the browser gets slow with 10+ tabs
  • Mental notes do not survive restarts — if it is not in a file, it is gone

Surprises

  • The agent is better at routine tasks than creative ones
  • Writing good prompts for sub-agents is harder than writing the code yourself
  • The agent catches things I miss (like checking spam folders)

Cost

Running this 24/7 costs roughly $0 in API fees. The real cost is the MacBook running as a server.

What is Next

  • GitHub Issues as a task queue
  • Voice morning digest via TTS
  • Auto-publishing pipeline

Try It Yourself

OpenClaw is open source: github.com/openclaw/openclaw

The learning curve is real, but once it clicks, you will wonder how you worked without it.


What automation have you built with AI agents? Drop a comment.

Top comments (0)