TL;DR
A framework that gives LLMs a terminal to control browsers and complete web tasks through code-based execution.
Key features
Terminal-based browser control: LLMs create and manage browser sessions via Playwright through the terminal.
Code-based task persistence: Web agent browsing history is stored as a single Python script for reuse.
Multi-backend support: Supports various LLM backends including OpenAI, Anthropic, and OpenRouter.
Plugin integration: Provides plugins for integration with other agents like Claude Code and Codex.
When to use it
When complex multi-step web automation tasks are needed
When you need to overcome limitations of existing web agents for more efficient browser interaction
When you want to store and manage web tasks as reusable code