Playwright Visual AI Monitor

Playwright Visual AI Monitor

Browser automation meets AI analysis: Playwright captures the page, the LLM inspects the screenshot for anomalies or changes, and your team gets a Slack ping if anything looks wrong.

What this workflow does

1. Cron trigger fires on schedule (e.g. every 30 minutes)
2. Playwright navigates to the target URL and takes a full-page screenshot
3. LLM analyses the image for visual anomalies, errors, or unexpected changes
4. Condition fires Slack only when the LLM flags an issue
5. Slack sends the alert; Output logs clean runs

Use cases

• Production dashboard uptime monitoring
• UI regression detection without test suites
• E-commerce price or availability tracking

Setup

Set your target URL in the Playwright node. Configure Slack credentials in Heym. Tune the LLM system instruction to describe what "normal" looks like for your page.

FAQ

Does this replace visual regression tests? It complements them — great for pages with dynamic content where pixel-diff tests produce too many false positives.

Which LLM supports image input? Use a vision-capable model (e.g. gpt-4o, claude-opus-4-7).

Can it survive a site redesign? Add a Playwright AI step with auto heal so broken selectors are repaired at run time. Self-Hosted AI Agent Templates: 6 You Can Watch Run shows auto heal in a recorded run.

To wake the vision model only when the page changed, add a Condition node in front of it. Our guide to gating a model behind a filter shows what that saves.