LaVague is an open source framework designed to help developers build AI-powered web agents capable of automating tasks across websites and web applications. It implements the concept of a Large Action Model framework, allowing agents to interpret a user-provided objective and translate it into a sequence of actions performed in a browser. These agents can navigate web pages, retrieve information, fill out forms, and execute multi-step workflows automatically. LaVague is centered around a World Model that analyzes the current webpage state and determines the next set of instructions, combined with an Action Engine that converts those instructions into executable automation code. It can use browser automation tools such as Selenium or Playwright to interact with websites programmatically. Developers can integrate various language models and configure the agent’s reasoning and execution behavior to suit different automation scenarios.
Features
- Framework for building AI web agents that automate browser tasks
- Converts natural language objectives into executable browser actions
- World Model component that analyzes page state and generates instructions
- Action Engine that compiles instructions into automation code
- Supports browser automation drivers such as Selenium or Playwright
- Includes tools for testing, debugging, token usage tracking, and demos