Getting started with verifAIed

verifAIed checks the work your coding agent says it finished: that the code it wrote is tested, and that the feature works in a real browser.

Prefer to watch? The tutorial video shows the everyday workflow from start to finish, then walks through setting up an account.
Your coding agent can answer questions about verifAIed. Once it is connected, ask it how something works: it reads this documentation through the docs tool. See MCP tools.

What verifAIed does

Your agent says it is done. verifAIed answers the two questions that leaves: what did it build, and is it really finished?

  1. Your agent writes the code and its tests, then uploads the test coverage to verifAIed.
  2. It records browser tests: flows through your running app, with a check after each step and a video of the result.
  3. It asks verifAIed check_done. The answer is done, or a list of gaps, each with the next thing to do.
  4. It fixes the gaps and asks again, until the answer is done.
  5. You open the branch and watch the videos, instead of checking the work by hand.

Build a feature with your agent walks through that loop once. New here? Start with Set up.

What verifAIed works with

  • Repositories on GitHub. verifAIed connects through a GitHub App; other hosts, such as GitLab and Bitbucket, are not supported.
  • Unit coverage for Python, from pytest with pytest-cov, and for JavaScript and TypeScript, from vitest with its Istanbul provider.
  • Browser tests for web apps, recorded and replayed with Playwright.
  • Claude Code, Codex, Cursor, Antigravity, and any other coding agent that can connect to an MCP server.

In this documentation

Get started

  • Set up: Connect GitHub, connect your coding agent and set up a repository.

Everyday workflow

Browser tests

Beyond tests

  • Code reached: Which functions a real browser actually ran.
  • Visual checks: A free audit of the rendered page, and a verdict on a screenshot.

Connect

Reference