lexmount/moli ? reverse-engineered prompt

Build me a fast, production ready headless browser for AI agents that works on Linux, macOS, and Windows.

I want it to be able to fetch web pages, extract useful text, and automate normal browser tasks without needing a full heavy browser running all the time. Most of the time it should work from the page structure first, then only do layout or screenshots when needed. It should support a CLI for things like getting Markdown, HTML, JSON, semantic text trees, screenshots, full page images, and PDFs.

I also want a local automation server that can be used through CDP, WebDriver Classic, and WebDriver BiDi, so tools like Playwright can connect easily. If you need to look up current docs online while wiring it up, go ahead. Make sure it feels polished, efficient, and ready for real use by scraping, crawling, and AI agent workflows.