Selenium WebDriver

[emsi]

I was thinking about adding a selenium WebDriver implementation. I've checked that GPT-4 is quite capable of interfacing with it. This could be a great, albeit slower, alternative for accessing the web.
In future we could use graphic output for GPT-4 multimodal capabilities.

What do you think?

[YaswanthDasamandam]

Yeah sounds good

[SlistInc]

That's a great idea, but have a look at the ongoing discussions on #4 and #121 as there is some work being done on the web-crawling functionalities there already. It would be a shame to duplicate efforts.

[BrandonBahret]

Using selenium in the "browse website" command is actually necessary for many sites as they use client side rendering with javascript. Not a "nice to have," rather you are wasting time and api calls using the current method of just looking at the raw html.

For example, if you wanted to find the price of the 4090 you would see "Starting at $XXX.XX" instead of "starting at $1599.99." Without using a web driver, you will get a lot of wasted api whenever AutoGPT could've found the answer sooner.

[BrandonBahret]

The best way of doing this would be to create a new plugin system for AutoGPT wherein people implement an interface, and then AutoGPT automatically updates the system prompt for the additional tools. Could have an entire repo just for plugins.

[SlistInc]

I agree, I am eagerly awaiting the plugin architecture being worked on. That will make sure this repo also doesn't bloat and Auto-GPT can stay nimble.

[GoMightyAlgorythmGo]

Would be fun to go to a website and chat/talk to it and show it interessting sources. But i think that is future stuff. At most you can copy paste the stuff and put it into a txt file and put it inside a folder and tell autoGPT to check it out or some predetermined folders like:
"userfeedback_folder" "high_priority_milestone" "error_user_copy_paste_folder -> then you can make a copy paste into a txt and next time AutoGPT has time it will check it out. But ja.... idk It definitly can read and write to txt files even if it cant it can write .py programms to do so!

[Merlinvt]

Does it make sense to integrate https://github.com/richardyc/Chrome-GPT ? I played around with it recently and it works quite well. Actually this fork worked even better https://github.com/BaseInfinity/SeleniumGPT

[Photon1c]

the read_webpage action in web_selenium.py throws an exception indicating that there are multiple "url" values being passed as arguments, btw. Seems that the agent is trying to read more than one web page.