Detect. Verify. Diagnose. Alert.
Sentinel watches your store, proves a problem is real with E2E tests, explains the likely cause with AI, and tells your team in Slack or email, before customers do.
From manual checking to a system that does it for you
Today QA checks several flows by hand after every deployment or incident. Sentinel makes the first layer of verification automatic.
Today
Someone notices a problem, then QA investigates by hand.
Something breaks, someone notices, QA investigates
With Sentinel
The system spots the change, reproduces it, and sends the evidence.
Something changes, system detects, E2E verifies, alert sent
How it works
Four steps, from a store signal to a report your team can act on.
Watch the signals
Sales, traffic and abandoned checkouts are monitored. An unusual drop triggers extra checks.
Run E2E tests
Playwright re-runs the critical journeys to confirm whether the issue is reproducible.
AI analysis
Test results plus anonymised session data go to Claude for an initial diagnostic report.
Tell the team
A clear message lands in Slack or email with the flow, evidence and next checks.
Two goals, one pipeline
An automated test pipeline first, then an AI layer on top.
1. Store health and E2E testing
- PDP to Add to Cart to Cart
- Collection Quick Add
- Search
- Cart functionality
- Scheduled runs plus a manual trigger after each deployment
2. AI-assisted diagnostics
- What failed, and which flow was affected?
- Is the issue reproducible?
- What is the most likely root cause?
- What should we check next?
- Session context from open-source rrweb, with strict privacy controls
The difference in one message
Not just "the checkout test failed". A report that saves the investigation time.
Sentinel verified the issue in 3 of 3 runs after the latest deployment.
- Affected flow
- PDP, Add to Cart, Cart
- Session behavior
- Add to Cart is clicked, no cart update follows
- Likely cause
- Cart request fails after the variant selector changes
- Check next
- Recent theme changes to the product form
Illustrative example only.
Start small, prove the value
A low-cost proof of concept first. Scale only if the results are good.
AI model access
Claude or another suitable model. Cost depends on the model and how often analysis runs.
Infrastructure
A small server and storage to host session replay and the data it needs.
Existing tools
Reuse current automation and Slack workflows. No new platform from day one.
Ready to move from reacting to detecting?
We need two approvals to begin: access to an AI model, and a small amount of server capacity.
Email us