# Swarm Tester

> A swarm of cheap AI agents drives real browsers against your app, hunting bugs from personas built on your real bug history, while one judge rules on what they find.

Swarm Tester dispatches a fleet of cheap-model agents to drive real browsers against your live app, each one hunting a specific bug drawn from your own bug history. One judge reviews the evidence and rules on every finding.

- Website: https://swarmtester.theportlandcompany.com/
- Get it: https://swarm-tester.theportlandcompany.com
- Availability: Invite-only during early access.
- Platform: Web
- Made by: Spencer Hill, The Portland Company (https://theportlandcompany.com)
- Privacy: https://theportlandcompany.com/privacy

## What you get

- **Real browsers, not scripts**: Each agent drives an actual headless browser against your target app, clicking, typing, and navigating the way a real tester would, not replaying a fixed script.
- **Personas from your own bug history**: Every check is an adversarial persona: a specific bug to hunt and a yes/no question for the judge, written from the kinds of bugs your app has actually shipped before.
- **A swarm, not one tester**: A sweep runs every persona across every role — UI, login, registration, SMS, email, accessibility, mobile, and more — in parallel, and reports only once every result is in.
- **One judge rules on every finding**: Agents only gather evidence. A single separate judge model reviews what each one saw and decides whether the bug is actually present, so verdicts are consistent.
- **Read-only by design**: Agents can navigate, search, and inspect, but every step is blocked at the network layer from sending, paying, inviting, or deleting anything — testing never touches live data.
- **Confirmed bugs land where you work**: Route confirmed findings straight into your issue tracker or any webhook you control, complete with the evidence the judge based its call on.

## How it works

1. **Point it at your app**: Register a target app and its environment, and Swarm Tester can suggest personas tailored to what it finds on your homepage and screens.
2. **Run a sweep**: Dispatch every persona across every role in parallel. Each agent drives a real browser, gathers evidence, and stops as soon as it has enough.
3. **Review the verdicts**: The judge rules on every finding. Confirmed bugs come with the evidence behind them, ready to deliver to your issue tracker or webhook.

## FAQ

### What kinds of bugs does it look for?

Whatever personas you have: UI rendering, login and registration flows, SMS and email sends, accessibility, mobile layouts, content and copy errors, and full user journeys, among others.

### Can it break or change my live app?

No. Every agent step is read-only, and writes like sending, paying, inviting, or deleting are blocked at the network layer regardless of what an agent attempts.

### Who decides if a bug is real?

A single judge model reviews the evidence every agent collected and rules on each finding, so the same standard is applied across the whole sweep.

### Can I write my own checks?

Yes. You can add personas by hand, or ask Swarm Tester to suggest new ones based on your app's screens and the coverage you already have.

### Where do confirmed bugs go?

You can deliver a confirmed finding to a connected issue tracker or to any webhook endpoint you control, along with the evidence that backs it.

### Does it support multiple apps or teams?

Yes. Targets, personas, and findings are scoped per organization, with roles controlling who can view results or make changes.

## More from The Portland Company

https://theportlandcompany.com/apps