LaVague
An open-source Python framework for AI web agents, with Selenium, Playwright and Chrome support. Uses OpenAI by default and carries an Apache 2.0 license.
LaVague is a Python framework for developers building AI agents that carry out tasks in a web browser. An agent takes a plain-language objective, examines the current page, and generates and executes browser actions across multiple steps. It's open source under the Apache 2.0 license.
Its design separates deciding what to do from carrying out browser actions. A World Model interprets the objective and page state, while an Action Engine turns its instructions into executable automation. Developers can customize the model configuration. Browser automation runs in your environment, but the default setup calls OpenAI's API and requires an API key and internet access; it isn't fully local by default.
LaVague works with Selenium, Playwright and a Chrome extension driver. Driver choice affects capabilities: Selenium supports headless operation and multiple tabs, while both Selenium and Playwright can handle iframes. An optional Gradio interface provides an interactive way to try agents. Testing and benchmarking tools help developers assess agent behavior, and logging, debugging and token counting help investigate failures and estimate model usage. For QA engineers, the companion LaVague QA tool turns Gherkin specifications into web tests.
Telemetry is enabled by default. It collects objectives, generated actions, visited URLs and retrieved HTML excerpts, along with model usage and error details. You can disable telemetry, though the default OpenAI model calls still use a cloud service.
