Building Hover: vibe-test your app, ship a real Playwright spec
How Hover drives your real Chrome to verify a flow you describe in plain English, then crystallizes it into a standard @playwright/test spec that runs in CI with zero AI.
End-to-end tests are where good intentions go to die. Writing them is tedious, maintaining them is worse, and one shifted selector turns the whole suite red for the wrong reason. Hover is our answer: vibe-test your app, ship a real Playwright spec.
Describe the flow in English
You don't write selectors. In the Hover panel inside VS Code, you type what you want checked:
log in and add a t-shirt to the cart
That's the whole input. No config, no bundler plugin, nothing added to your app.
AI explores once, in a real browser
Hover connects to your real Chrome over the Chrome DevTools Protocol, and an agent runs the flow the way a person would, reading the live page instead of guessing at selectors. When the run finishes clean, it has a verified path through your app.
The agent runs on the Claude or Codex CLI already on your machine, so there's no API key to wire up and no token bill from us.
Then it crystallizes
This is the move that matters. Hover doesn't keep the AI in the loop. One click on Save as Spec writes a plain @playwright/test file into your repo:
tests/add-to-cart.spec.ts
After that, your CI runs standard Playwright. No API keys in CI, no token bill, no model having a bad day and failing a green build. The test is yours, and it keeps working whether or not any model is online.
That's the whole philosophy in one tool. AI authors it, you own and run it. Hover is open-source and local-first. Try it at gethover.dev.
Related Articles
The harness is the product
When every team can call the same frontier model, the model stops being the moat. The engineering that matters moves to the harness: the tools, grounding, and verification around the agent.
AI authors it. You own and run it.
Let AI do the expensive exploratory work once, then crystallize it into a durable artifact that runs forever without the AI tax. The pattern behind everything we build.
AI-adjacent vs. AI-native: how to tell the difference
Most AI features are a chatbot stapled to software built before agents existed. AI-native means the whole thing assumes an agent from line one. Here is a checklist that tells the two apart.