LaVague SOP

Appearance

LaVague ​

Category: OSS Agent

Free / Paid: Free and open-source (Apache-2.0 license). LLM provider costs apply -- defaults to OpenAI gpt-4o, so a valid OPENAI_API_KEY is required at runtime.

Official site: https://lavague.ai/

GitHub: https://github.com/lavague-ai/LaVague (5,500+ stars)

Docs: https://docs.lavague.ai/

What it is ​

LaVague is an open-source Python framework for building web agents powered by Large Action Models. You write a natural-language objective and the agent navigates, clicks, and fills forms inside a real browser to complete it.

The framework has three core components:

  • WorldModel -- reads the current page context and plans the next action
  • ActionEngine -- translates the plan into executable browser commands (Selenium by default)
  • WebAgent -- orchestrates WorldModel and ActionEngine in a loop until the objective is met

LaVague ships a Gradio-based interactive demo interface (agent.demo()) so you can watch the agent work step by step and approve or reject each action before it executes.

Requirements ​

  • Python 3.10 or higher
  • A Selenium-compatible browser driver (ChromeDriver installed and on $PATH)
  • An OPENAI_API_KEY environment variable set to a valid OpenAI key

Safety and telemetry note ​

LaVague collects usage telemetry by default, including your agent objectives and page interactions. To opt out, set the environment variable:

bash
export LAVAGUE_TELEMETRY=NONE

The docs also warn: never include personal information in your objectives, since objectives may be logged.

The agent operates a real browser and can click, fill forms, and submit data. Review every objective carefully before running against live accounts or production sites.