Skip to main content

From Zero to One: How a Solo Developer Built an Intelligent Ops Assistant with YingClaw

Li Ran (pseudonym) is a solo developer simultaneously maintaining 3 SaaS products and 5 VPS instances. His daily routine? Being chased by server alerts.

The Old Reality​

  • 5 VPS instances (across Alibaba Cloud, Tencent Cloud, and AWS Lightsail)
  • ~20 monitoring alerts per day
  • Most alerts were non-urgent issues like "disk usage at 85%" — needing cleanup but not emergency response
  • Truly critical alerts got buried in the noise

"One night I got a CPU alert at 2 AM. I figured it was normal fluctuation, rolled over, and went back to sleep. Turned out it was an infinite loop that crashed the service. Six hours of downtime."

The Solution​

He built an Ops Agent with YingClaw, running on a low-spec VPS (1C2G, ¥40/month).

The architecture is straightforward:

Cron job (every 15 minutes)
  → Agent fetches monitoring APIs from all servers
  → Agent analyzes alert severity
  → Critical: WeChat notification + attempt auto-remediation
  → Non-critical: Logged, included in daily summary

Real Results​

Scenario 1: Automatic Disk Cleanup​

Agent detects /var/log at 85% usage
  → SSH into target server
  → Run logrotate
  → Free 12 GB
  → WeChat notification: "Auto-cleaned ✓"

30 seconds end-to-end. Zero human intervention needed.

Scenario 2: SSL Certificate Expiry Warning​

Agent detects certificate expiring in 7 days
  → Auto-renew via acme.sh
  → Verify new certificate
  → WeChat notification: "Renewed ✓, new expiry: 2026-08-25"

Scenario 3: Abnormal Traffic Detection​

Agent detects 10x QPS spike on an API endpoint
  → Analyze traffic sources
  → Identify as scraping attack
  → Auto-add iptables rules to block IPs
  → WeChat notification: "Auto-blocked 5 malicious IPs ✓"

ROI Summary​

MetricValue
Setup time2 days (including skill authoring and debugging)
Monthly cost¥40 (VPS) + ¥15 (API fees)
Alert handling timeFrom 2 hours/day to 10 minutes/day
Downtime incidentsFrom 3/month to 0/month
Sleep quality"I finally get uninterrupted sleep"

Key Takeaways​

1. The barrier to AI ops is lower than you think​

You don't need to be an AI expert. Shell scripting skills + understanding your server metrics is enough.

2. Progressive trust​

For the first three weeks, the Agent only did "detect + notify." Once its judgment proved accurate, auto-remediation was gradually enabled.

3. Notifications matter​

No matter how much the Agent does, if it doesn't tell you, you won't trust it. Every action pushes a WeChat notification. Building trust beats showing off.

"Now I glance at WeChat every morning and last night's ops report is already in my chat history. It feels great." — Li Ran