How do I verify code my AI agent just wrote?

Run qagent --url " ". QAgent drives the browser and returns PASS/FAIL with evidence — pipe --reporter=ndjson straight back to your coding agent.

Is QAgent open source?

Yes — MIT-licensed on GitHub, published as @qagent/cli on npm. You pay only for the LLM tokens it uses.

Does QAgent replace Playwright or a GUI testing SaaS?

No. Playwright remains useful for long-lived regression suites, and GUI platforms like Mira Checks cover continuous no-code monitoring. QAgent is the fast terminal-side check for the development loop.

What does an AI browser testing CLI cost?

The CLI is MIT-licensed and free — you pay only for the LLM tokens a run uses. In the README benchmark, a multi-field form check came in around $0.008 with google/gemma-4-26b-a4b-it and around $0.05 with gpt-4.1-mini, both 5/5.

Open Source · QAgent CLI

AI browser testing CLI: verify websites from the terminal

Automation builds. Verification proves.

Testing platforms are launching their own "best AI browser testing CLI" pages — usually the same recording and automation platform with a terminal veneer. This is the verification-first take: a CLI that actually runs, checks your page in a real browser after every change, and hands the verdict back to you or your coding agent. QAgent is an open-source AI browser testing CLI built for exactly that.

Try it on your site How to test after every deployment

Real CLI interaction

One command, a real browser, a verdict

Install, point it at your app, describe what should work. A driver model picks the browser actions, Playwright executes them, and a separate verifier model returns a structured PASS/FAIL with evidence.

The verdict is scriptable: stable exit codes (0 pass · 1 fail · 2 config · 3 runtime) and --reporter=ndjson events mean your coding agent or a script can react to the outcome.

Reporters: list · json · ndjson · trace Composable with commas MIT · Pre-1.0

# Install

npm install -g @qagent/cli
npx playwright install chromium

# Verify a page after a change — from the terminal

qagent --url http://localhost:3000 "Submit the contact form and check the success message"

Cost in the README benchmark: a multi-field form check ran for as little as ~$0.008 with google/gemma-4-26b-a4b-it (5/5) and ~$0.05 with gpt-4.1-mini (5/5). The CLI itself is free — you pay only for the LLM tokens.

See it run: qagent.mirachecks.com · npm · GitHub

Verification vs automation: the split that matters

Most "AI browser testing CLI" pages are vendors repackaging their automation platform. The distinction that actually matters when you're building:

Automation

Maintains a suite of defined tests and runs them on a schedule. Catches regressions you already anticipated.

Verification

Answers "does the page actually work right now" — after each change, in a real browser, from the terminal. Catches the thing you just broke.

AI coding made this urgent in a way test suites never were. When a coding agent writes a feature in minutes, the fastest honest check is a goal in plain English and a real browser run — not a new test harness the same agent just generated. That is the verification loop, and it belongs in the terminal next to your agent, not behind a SaaS dashboard.

For the full workflow around a release, read how to test your website after every deployment.

When to reach for the CLI, the GUI, or Playwright

These are complements, not rivals. Each covers a different phase of the same problem.

QAgent — terminal verification

The fast inner-loop check. You (or your coding agent) run a goal, get a verdict with evidence, and keep going. No dashboard, no test authoring.

Try QAgent

Mira Checks — GUI platform

Autonomous, no-code AI website testing with continuous monitoring and visual regression. For teams that want coverage without running anything from a terminal.

Open app.mirachecks.com

Playwright — regression suites

Right for long-lived regression suites your team deliberately maintains. QAgent uses Playwright under the hood and stays out of its way — same browser engine, different job.

See the verification loop in practice

Frequently asked questions

How do I verify code my AI agent just wrote?: Run qagent --url <your-app> "<what should work>". QAgent drives the browser and returns PASS/FAIL with evidence — pipe --reporter=ndjson straight back to your coding agent.
Is QAgent open source?: Yes — MIT-licensed on GitHub, published as @qagent/cli on npm. You pay only for the LLM tokens it uses.
Does QAgent replace Playwright or a GUI testing SaaS?: No. Playwright remains useful for long-lived regression suites, and GUI platforms like Mira Checks cover continuous no-code monitoring. QAgent is the fast terminal-side check for the development loop.
What does an AI browser testing CLI cost?: The CLI is MIT-licensed and free — you pay only for the LLM tokens a run uses. In the README benchmark, a multi-field form check came in around $0.008 with google/gemma-4-26b-a4b-it and around $0.05 with gpt-4.1-mini, both 5/5.

Try it on your site

Open-source, installable in two commands. Prefer a GUI? Start with the no-code platform instead.

Try QAgent on your site

Also read: testing AI-generated code · post-deployment testing guide · the best AI website testing tools in 2026