Overview

Node Mantis lets your software run an AI agent that uses the web. The browser capture and actions run on your machine. Task instructions, selected page and DOM context, screenshots, and action history are sent to the hosted service when needed for decisions and retrieval. The service also provides account authentication and usage-based billing.

A browser-agent turn follows this flow:

1. Your client captures the current page (DOM + screenshot).
2. It sends selected context to the service, which returns the next action.
3. Your client executes the action locally, then repeats.

The distribution is published as node-mantis on npm and PyPI. Both packages expose the browser automation SDK — TypeScript with async methods and camelCase options, Python with blocking methods and snake_case options — backed by the same platform-selected compiled runtime. The npm package also contains the nodemantis command line, which runs goals and retrieves data straight from a terminal. Python additionally ships the configuration, platform-check, browser-install, and installation-smoke tools; attaching to a caller-owned Playwright page remains TypeScript-only.

Create an account, open the dashboard, and generate an API key before calling Node Mantis.

1. Go to Sign up and create a Node Mantis account.
2. Open the dashboard and find the API keys section.
3. Create a key, copy it when it is shown, and store it as an environment variable or in the Python config file.

export NODEMANTIS_API_KEY="sk_live_<copy-from-dashboard>"

Run this command in the terminal that will start your application. It makes the key available to programs launched by the current shell and is cleared when that shell closes. Anyone with the key can use your account and consume its usage balance, so keep it out of source code, browser bundles, logs, and built artifacts. See API key setup for PowerShell setup.

TypeScript can also receive the key through the apiKey option. Python can store it with python -m nodemantis configure --api-key "$NODEMANTIS_API_KEY".

New here? Start with the Quickstart, or read How it works for the mental model.

• Quickstart — key, install, first run.
• Concepts — the perceive → decide → act loop.
• API reference — SDK, CLI, config, and result contracts.
• CLI — run and configure the agent from the terminal.
• Examples — task recipes.