PerfAgents
A synthetic monitoring platform. You brought the browser tests you already had, and we ran them as monitors from around the world, every few minutes, and woke up the right person when something broke.
- Founder
- 2020 to 2025
- About 20 people at peak, US and India
- Acquired by Testing Mavens, April 2025
What it did
Most teams already had Playwright, Cypress, Selenium or Puppeteer scripts for testing. PerfAgents ran those same scripts on a schedule, from AWS regions around the world, against production. Not a ping to the homepage. The real user path, hitting the real APIs.
When a step got slow or failed, the alert went to the engineer who owned it, before customers noticed. Scripts could live in a team's GitHub repo, and runs could be triggered from CI/CD through the API.
Who used it
One of our customers was a fintech whose components sat inside the checkout flow of major US retailers such as Nordstrom and Lululemon. Our nodes walked that checkout path from Frankfurt, Singapore and Virginia, every few minutes.
How it was built
The platform, PerfAgents Uncloud, was multi-tenant. Each customer got their own subdomain, routed by a load balancer to an environment of its own. Most services were Node.js and TypeScript, with some Java and Spring Boot. I wrote the core orchestration service and chose Lambda or Kubernetes for each workload, with KEDA and Karpenter scaling the cluster. It was listed on AWS Marketplace.
At its peak it ran about 10 million monitor executions a month at 99.99% availability.
The team and the exit
We bootstrapped it, with no outside funding, and grew to about 20 people across the US and India. Testing Mavens acquired the platform in April 2025.
What I learned
Enterprises don't buy features. They buy the ability to sleep at night. Not once did a feature list close a deal. What closed deals was a fast, convincing answer to one question: what happens when your thing breaks at 3am.