Implement AutoGPT

[0x4D5352]

Currently, the prototype manually deploys drones to a test hive - a linux server that it can use SSH to connect to as root using an RSA public key to minimize user interaction. This could and should be made more varied in order to finish the prototype.

The current goal for successful implementation is if AutoGPT can use SSH to connect from localhost to a target machine on the local network.

[0x4D5352]

Got AutoGPT4ALL running, but (apart from the non-chatGPT model performing poorly) I'm mainly running into JSON based errors. Changing to a different memory type might be useful, but should do more debugging first. Test GPT-4 with non autogpt4all setup to rule out issues with your computer!

[0x4D5352]

TODO: update .env.template for autogpt4all setup to match the newer .env.template in auto-gpt that supports local models more easily

[0x4D5352]

Good news: OpenAI implementation works! Bad news: the Auto-GPT master template breaks stable so autogpt4all doesn't work properly.

Screenshot below of GPT3.5 turbo using SSH to connect to a server on my local network and using the ip addr command to get network information, as a proof of concept that this can work. It cost roughly $0.10 to get to the point in the screenshot. The AI had to spend time learning how to actually run these commands, and didn't quite understand that it had actually completed the task, showing that more effective initial prompting can reduce overhead costs if this is the direction that is chosen for the full project or as an alternate path for people who don't mind connecting to OpenAI.

Leaving this issue open because while i've hit PoC for this, I need to actually implement it as a standard prompt so that the user doesn't have to specifically type it in for the demo.