hal-9100
hal-9100 copied to clipboard
code interpreter: pack an headless browser / browser mode or tool
basically LLM should be able to do this:
"yo here's a website https:abcd.com can you do x and y and get w and q from it plz?"
code interpreter would gen python code to do these stuff in headless browser step by step (including doing clicks, filling forms, etc. using playwright api for example) and then go back to the user and user happy
just gotta:
- edit docker image by adding playwright (they have a docker image imho)
- maybe tweak prompt
- #25 is prob mandatory to do before
also i think code interpreter should have some options for ppl that dont want browser for example
(openai by default block scrapping in their code interpreter for security reason i assume)