Asked and answered.

No. KaleFlow is local-first. The CLI runs the crawl and the replays on your machine, stores the evidence there, and opens the canvas in your own browser.

The hosted app at app.kaleflow.com is optional. A run reaches it only when you choose to upload that run.

KaleFlow works from two inputs: a code repository and a live URL. The crawl and the replays drive your app through a browser, the way a user would, so they do not depend on how the app is built.

The Repo Brief reads your repository to predict flows. If your stack is unusual, run the three commands against it and read the brief: it lists the flows it found and where in the code it found them.

The AI works once, during the crawl. What it produces is a compiled replay: a fixed list of steps for one flow.

Re-running a replay executes those steps in the same order. There is no model in the loop, so there is nothing to improvise. If a replay that passed before now fails, your app changed.

You don't write them. The crawler explores your live app and compiles the replays for you, starting from the flows the Repo Brief predicted.

And every replayed step carries evidence: a screenshot, video, a trace, a HAR file and the console output. A hand-written test tells you an assertion failed. A replay shows you the screen where it failed.

A path through your product that ends in something a user came for: an account created, a payment made, a teammate invited. These are the flows KaleFlow maps and replays.

Yes. Some flows only complete when one user hands something to another, such as an invite that a second person has to accept. KaleFlow runs both sides and shows the hand-off on the canvas.

Screenshots, video, a trace, a HAR file of the network requests, and the console output. Passing steps keep their evidence too, not only failures.

Run three commands in your repository: npx kaleflow init --app <your URL> --repo . then npx kaleflow crawl, then npx kaleflow ui. The last one opens the canvas.