adi-jagan/globot

Globot is an agent that controls your browser using playwright and GPT-4V.

globot

With GPT-4V, we can finally complete the original vision of natbot.

Help solve general agents by contributing to this repo!

Ideas for Improvement

• Scrolling (easy to add, but likely to cause divergence)
• Better context management (learning from mistakes, more descriptive history)
• Masking the image with node IDs
• Better DOM parsing (please submit issues/PRs!)
• More explicit planning
• Data collection and fine-tuning

NOTE: Remember to use the latest release of the openai API for the vision model:

pip install --upgrade openai

---

made by Ivan Yevenko