Tensteps

Tensteps sends AI personas through your site's preview deploys and reports, on the pull request, where people got stuck.

Each persona is a different person: a retiree on an old Android phone, a skimmer in a hurry, a screen-reader user. They're given a goal from your tensteps.yml ("buy any pair of shoes under $50") and they try to reach it the way that person would, on that device and connection. You get back who made it, where the others gave up, and a step-by-step replay of everything they saw, thought and tried.

Tensteps is in a pilot. The CLI ships on npm as tensteps@beta; plain tensteps is a placeholder until 0.1.0. Join the pilot and we'll set it up with you.

How it fits your workflow

1. You open a pull request. Vercel builds a preview.
2. Tensteps runs when the preview is live. A GitHub Actions workflow starts npx tensteps@beta run, which reads the preview URL and the pull request from the deploy event.
3. Results land on the PR as one comment that updates on every push. The full HTML report, with screenshots of every step, is attached to the workflow run.

Where to start

• Quickstart: run it against any URL from your laptop in five minutes.
• GitHub + Vercel: test every pull request's preview automatically.
• tensteps.yml: every option in the config file.
• CLI: commands and flags.
• FAQ: cost, data, safety.

What it is not

Personas are simulated users, not real ones. Treat every finding as a lead to confirm, and the replay as the evidence. Tensteps is good at finding where a flow got harder to finish; it doesn't replace analytics from real users.