LaVague

LaVague is an open-source framework designed for developers who want to create AI Web Agents to automate processes for their end users. The agents can take an objective, such as "Print installation steps for Hugging Face's Diffusers library," and generate and perform the actions required to achieve the objective. The framework consists of a World Model that interprets objectives and page context into step-by-step instructions, and an Action Engine that compiles those instructions into executable code like Selenium or Playwright commands. Targeted primarily at developers and QA engineers, LaVague lowers the barrier to creating reliable automation scripts. Instead of writing brittle selectors or page-specific logic, users define high-level goals, and the agent handles action generation and execution. The framework also includes LaVague QA, a specialized tool that converts Gherkin specifications into automated tests, making it a natural fit for testing workflows. Key features include built-in contexts for easy configuration, a test runner for benchmarking, a TokenCounter for estimating costs, logging tools, an optional Gradio interface, debugging tools, and a Chrome Extension. LaVague supports multiple drivers: Selenium, Playwright, and Chrome Extension drivers, with varying support for headless agents, iframes, and multiple tabs. The framework uses LLMs under the hood (default is GPT-4o, fully customizable), and costs depend on the models chosen, objective complexity, and website. Users can track token usage and estimate costs with the TokenCounter tool. Compared to alternatives like Playwright or Puppeteer scripts, LaVague offers a higher-level abstraction that adapts to page changes, though it requires hands-on setup and is best suited for developers comfortable with Python and automation concepts.