, an agentic testing engine.

Find a path.
Run it again.

Describe a browser task. Cairn uses AI to find the steps, saves them as JSON, and runs them again without a model call.

Basic replay makes no model calls.

A browser task,
step by step.

One task, three views. Follow the sample app, the action being executed, and the path it leaves behind.

Step 1 of 4

Tell Cairn what to do. Watch it log in, add a daypack, and open the cart. Each action becomes a step you can save.

Simulated app and actions. No external site is opened.

Try this in your terminal

cairn discover "log in, add a daypack, and open the cart" \
  --url=https://your.app --freeze=cart.skill.json

Sample appReady to run

trailsupplyGuest

Action traceIllustrative code

1click("Log in")ready2click("Add to bag")3click("Cart")

Press play to follow the task.

Run it from a terminal, wire it into CI, or embed it in your own tool. Anything that varies comes in through a port. Explore the engine API

Pipeline

Driver — Drives the browser and waits for the page to settle. Chrome DevTools ships; bring Playwright or your own.

OutputRun a frozen scenario through the pipeline.

Give it a task.

$ npm install -g cairn-engine

Start with the guide Example project Discovery, repair, and AI checks may call a model.