Run it yourself

When to run it yourself

• Your product only runs on your computer. The hosted console cannot reach localhost. A copy running on your machine can.
• You want the evidence on your own disk. Findings, screenshots and verdicts stay in the folder you run it from.
• You want to use your own AI key. The checks that use AI run on a key you set, billed to you by that provider and never metered by us.

What you need

• A copy of the nth Labs repository. It is private, so write to hello@nthlabs.dev first.
• Node 20 or newer. Check with node -v. There is no build step: Node runs the code as it is.
• A model key, but only for the checks that use AI. The free checks need none.

Install it, in order

1. Install the dependencies

   npm install

   Run it in the folder you copied. It fetches the library that drives the browser, but not the browser itself. That is the next step.

2. Install the browser

   npx playwright install chromium

   This step is easy to miss. Without it, every check that opens a page fails at the start with a message about a missing executable.

3. Point it at your product

   node cli.cjs init

   It asks a few questions: the product’s name and address, what it is and who it is for, the pages to test (add | auth to a page behind your login), competitors, where the code lives, an optional test account and an optional model key. Only a name or an address is required. It saves your answers in files only your user account can read, then checks the setup against your real product in one browser pass.

4. Check the setup whenever you change it

   node cli.cjs doctor

   It makes one browser pass and spends nothing. It loads every page, tries the sign-in if there is a test account, and exits with an error when something is broken, so it is safe to put in front of anything else.

5. Run the free check, or open the console

   node cli.cjs checkup
   node cli.cjs ui

   checkup runs every free check in one pass and prints one summary. ui starts the console on your own machine, reachable only from it. It is the same console, minus the settings only a hosted one needs, such as plans and teams.

Adding a model key

Site review, User simulation and the simulated customers need a model key; everything else runs without one. Give a key to init, or add it later in the console under Settings → API keys, where it is stored only on your machine. Commands you type in a terminal read the key from your environment instead.

node cli.cjs models lists every model you can run, its price and whether its key is set, and when a key is missing it shows the variable to set. Runs on your own key are billed to you by the provider and never metered or capped by us.

Hosting a console for your team

The console can also run on a server of your own, so a team opens a web address instead of a terminal. It needs a long-running container with a disk for its data, not a serverless function, because a run drives a browser for minutes at a time. On an address other people can reach, it refuses to start until it has an owner account, set with LABS_EMAIL and LABS_PASSWORD. Like the hosted console, it will not open addresses on its own network, localhost included. docs/DEPLOY.md in the repository walks through it step by step.

Where to go next

• The command line: every check as a command.
• Claude, Codex, Cursor: node cli.cjs mcp lets your coding agent call the checks itself, as tools named labs_*.
• Running it in CI: check every preview deploy from your workflow.