moli: Lightweight Headless Browser for AI Agents

A production-grade headless browser optimized for AI agent tasks. It provides a complete browser runtime—including the V8 engine, DOM, CSS, and networking—while designing layout and rendering as on-demand processes that execute only when necessary.

1 / 5

The design is based on the insight that most automation tasks focus on extracting page structure rather than visual rendering. Layout calculations are triggered only when screenshots or coordinate inputs are required, minimizing resource usage otherwise to improve the efficiency of crawling and search pipelines.

It supports three protocols—CDP, WebDriver Classic, and WebDriver BiDi—in a single binary. It is compatible with existing tools such as Playwright and Selenium, and can be integrated immediately via CLI or server mode without installing separate drivers.

It supports various output formats, including HTML, Markdown, JSON, and semantic trees, allowing you to instantly obtain data processed for LLM input. It supports Linux, macOS, and Windows, making it suitable for agent-based web navigation and reinforcement learning environment setup.