auto-browser

Why it's ranked here

The core workflow is approachable and unusually broad, covering extraction, forms, multi-page tasks, templates, and reports through one CLI. The implementation weakens that promise: AI template analysis is discarded in favor of fixed finance-site selectors, so generated templates are not genuinely site-aware.

What's good

It closes the loop from task description to saved output. Browser sessions always close after work, templates use validated configuration models, existing template files can be extended, and Docker packages the browser-heavy dependency stack. Verbose output exposes generated selectors, raw extraction results, and processed content for debugging.

Tradeoffs

An OpenAI API key is mandatory, and browser tasks depend on model output rather than deterministic scripts. Interactive work may run a second agent over the same browser, increasing model usage. Headless mode is forced despite a documented environment setting. Installation guidance also conflicts with package metadata about the minimum supported Python version.

How to use it well

Use it for exploratory extraction, recurring site tasks, and browser workflows where natural-language convenience matters more than strict reproducibility. Prefer Docker, inspect verbose output, and hand-correct saved selectors before relying on templates. It does not replace deterministic browser tests, a general automation library, or a robust template generator in its current form.

Observed

License

MIT

Primary language

Python

Interface

Command-line interface built with Click

Packaging

Setuptools package with an auto-browser script entry surface

Installation

Docker Compose, Docker image build, shell installer, or editable pip install

Browser engine

browser-use with a headless browser and vision-enabled OpenAI model agent

Output

Markdown files, with optional model-generated structured reports

Platform support

Docker is described as cross-platform; the quick installer targets Linux and macOS