# How Forge QA is built

A Rust backend, Postgres that separates workspaces in the database itself, Playwright in disposable containers, and a network guard on every run. Here is what that means for your tests and your data.

## What happens when a test runs

1. **Trigger** A schedule, a CI trigger or a click on Run.
2. **API** Rust and axum. Checks the plan, reserves the run and queues it.
3. **Container** A fresh, disposable container with Playwright and the browser you chose.
4. **Network guard** Lets the browser reach only the origins approved for this website.blocked: private addresses, unapproved origins
5. **Your website** The test runs against your site. Results, evidence and security findings go back to your workspace.

## The parts, and why we chose them

- **Rust and axum**: The API and billing logic are written in Rust on the axum web framework. Fast, memory-safe, and strict about types where money and permissions are involved.
- **Postgres with row-level security**: Every workspace's data is separated by row-level security in the database itself, not only in application code.
- **Playwright**: Playwright drives Chromium, Google Chrome, Microsoft Edge, Firefox and Safari (WebKit). Traces come straight from Playwright, so you can open them in its trace viewer.
- **Safari (WebKit), not the Safari app**: WebKit is the engine inside Safari and catches most Safari-specific issues. It is not Apple's Safari app on a Mac or iPhone, and we don't claim it is.
- **One container per run**: Each run starts in its own disposable container, so no cookies, storage or state carry over from another run or another customer.
- **A network guard per run**: The browser can reach only the origins approved for that website. Private and internal network addresses are blocked, so a test can't be turned against internal systems.
- **Secrets stay secret**: Test secrets are stored encrypted and referenced by name. Passwords and secrets are masked in screenshots and logs.
- **Security scan in every plan**: Deterministic checks run during every test; AI only gives a second opinion on new third-party scripts. [Details](https://forgeqa.io/security-scan)
- **Evidence with an expiry**: Screenshots, video, traces and logs are deleted after 7 days on the trial and 14 to 60 days on paid plans. Run history stays, so you keep the record without the bulk.

## AI through OpenRouter

Test writing, ForgeAI review, AI Buddy, self-healing, the assistant and the scan's second opinion all use AI. All of it goes through OpenRouter.

- **The right model for each job.** OpenRouter lets us choose the model per task, and switch when a better one appears.
- **No lock-in to one AI vendor.** Not for us, and not for you.
- **Bring your own key.** On Agency you can connect your own OpenRouter key and choose which of the supported models it uses.
- **What the model sees.** The prompt and the page content needed for the task. For the security review, only short excerpts of new scripts.

## Services we rely on

- **Hetzner** hosts everything in its Falkenstein data center in Germany: the website, the app, your data and the test browsers.
- **Cloudflare** sits in front of forgeqa.io and the app to absorb attacks.
- **Stripe** for payments and for identity verification. Stripe handles ID documents; we don't store ID images.
- **OpenRouter** for AI models.
- **Resend** for account and notification email.
- **Slack and Atlassian (Jira Cloud)** only if you connect them.

## See it run against your site.

7 days, 100 runs, no credit card.

---

Source: https://forgeqa.io/technology · Start a free trial: https://app.forgeqa.io/#signup

Summary: How Forge QA is built: Rust and Postgres with row-level security, Playwright in disposable containers behind a per-run network guard, masked secrets, and AI through OpenRouter.