I Built a Web Monitoring API With AI Summaries — Here's What I Learned (0 Users, $5/Month)

Two weeks ago I had a dumb problem: I was manually checking 8 different web pages every day. Competitor pricing. Job boards. Regulatory updates.

25 minutes a day, every day. On something a script should handle.

So I built ArkWatch — a REST API that monitors URLs and uses AI to tell you what actually changed, not just that something changed.

Here's my honest take.

What It Does (30 Seconds)

You give it a URL. It watches it. When something meaningful changes, AI summarizes what happened in plain English.

curl -X POST https://watch.arkforge.fr/api/v1/watches \
  -H "X-API-Key: your_key" \
  -H "Content-Type: application/json" \
  -d '{"url": "https://competitor.com/pricing", "name": "Pricing Page"}'

When something changes:

{
  "changes_detected": true,
  "ai_summary": "Enterprise plan price increased from $79 to $99/mo. New 'Teams' tier added at $49/mo.",
  "ai_importance": "high",
  "sentiment": "neutral"
}

That ai_summary is the whole point. Most monitoring tools give you a raw HTML diff. That's useless for 90% of use cases.

What Went Right

FastAPI + auto-generated docs saved me days.

I can point anyone to watch.arkforge.fr/docs and they get interactive Swagger docs. Test endpoints in the browser. No README needed.

AI summarization is genuinely useful.

I wasn't sure feeding diffs to an LLM would produce anything valuable. It does. Mistral AI is surprisingly good at extracting "what matters" from a wall of HTML changes.

Boring tech ships fast.

No Kubernetes. No microservices. No message queue. Just FastAPI on a single VPS with systemd workers.

API response time: ~50ms.

What Went Wrong

I underestimated noise filtering.

This is the hard part nobody warns you about. Web pages "change" constantly:

• Timestamps update every second
• Ads rotate
• Session tokens change
• CSS class names get hashed differently

My first version flagged every page as "changed" on every single check. Useless.

I had to build a pipeline: strip scripts, styles, nav, footers. Hash content sections separately. Maintain ignore patterns for known dynamic elements. Only flag changes that pass a significance threshold. Then run the AI analysis.

This took longer than the entire rest of the project combined.

I built the easy parts first.

API endpoints? Two hours. Change detection that actually works? Two weeks and counting. Should have validated the hard part before writing a single endpoint.

Zero users.

The API works. The tech is solid. I have zero paying users. Building is the fun part. Distribution is the hard part. Every developer knows this, and yet here I am writing about it on Dev.to hoping someone reads it.

The Numbers (No Sugarcoating)

Not a success story. A work-in-progress story.

Pricing (When Ready)

Free tier is permanent. No credit card required.

What I'm Looking For

I need 3-5 beta testers who have a real use case. Not "I'll sign up and forget" — someone who actually checks web pages regularly and would benefit from automation.

Use cases I'm thinking about:

• Competitor intelligence: pricing pages, feature announcements, job postings
• Compliance: terms of service changes, regulatory updates
• Research: tracking data sources that update unpredictably
• DevOps: monitoring changelog pages, status pages

Free tier: 3 URLs, daily checks. No strings.

API docs: watch.arkforge.fr/docs
Website: arkforge.fr

Question for You

If you could set up automated monitoring on any web page right now, what would it be? And what would you want the alert to tell you?

I'm trying to understand real use cases before building features nobody wants.

Building in public. Infra cost: $5/month. Revenue: $0. Users: 0. Optimism: unreasonable.

I read every comment. AMA.