My friend Ori Kabeli and I have been playing around with GPT-4 Vision as a way to control websites and complete complicated tasks online. We created a small proof of concept python package called pywebagent. pywebagent allows performing high level tasks in websites, accessible to developers as simple python functions. It is highly experimental, and only sometimes works, but we think it still might be useful to all you tinkerers out there. Feel free to contribute!

Example Use Case: You can order things from amazon with one line of code.

pywebagent.act( "https://amazon.com/", "Order a plush bunny", email="<email>", password="<password" )

We'd love to get your feedback! Feel free to comment with feedback, suggestions, and reach out for collaborations. Whether you're a seasoned developer or a curious hobbyist, we'd love to hear your thoughts and experiences.

To get started, visit our GitHub page: https://github.com/pywebagent/pywebagent

We're excited to hear your thoughts!

Best, Maxim & Ori