First Build

Before production this likely needs a lot more checks on output. I was thinking of running a second LLM pass just to see if it was dangerous code and how dangerous it was. Like if it tries to send rm -rf / then make it fail in all circumstances and make the user manually do the code themselves. While allowing "echo hi" to be sent. However, adding another LLM pass also increases costs and complexity.

So for now, let's just say this is for fun.

ALSO: I added an unsafe option that lets you have GPT-3 / GPT-4 just write a command and directly execute it... with no checking. It's literally madness. The only reason I included it was because I want to use it in a VM to play with. Seriously don't use it. I probably shouldn't include it at all but I also hate programs that omit good functionality just because the dev doesn't trust their users.

You have been given great power. Be responsible.