LaVague is an open-source Large Action Model framework for building AI web agents that can navigate websites and automate browser-based workflows. It uses a World Model (LLM-based) to interpret objectives and an Action Engine to compile them into Selenium or Playwright code. The framework is designed for developers building automation over SaaS tools (CRM, ERP, etc.) and supports local models like Llama 3 as well as cloud models. Note: the GitHub repository last committed in January 2025, indicating the project may be in low-maintenance mode. Key features: - World Model + Action Engine architecture: LLM plans, Playwright/Selenium executes - Supports OpenAI GPT-4o by default, configurable to local models (Llama 3, etc.) - LaVague QA: automated test generation from Gherkin specs - SaaS workflow automation (CRM, ERP data entry and extraction) - Fully open source, available on GitHub - Integrates with Hugging Face model hub
Free and open source. No commercial pricing.