Another attempt at building an AutoGPT-like thing based on my first project go-auto-gpt.
Instead of using the common plan and task pattern, FlowGPT will attempt to maintain internal dialogue to determine the correct steps to solve a problem.
FlowGPT uses a finite state machine to determine the next step in the overall workflow to solving a problem. It consists of three different internal components: thought, critique and action. Each component will maintain their own memory and prompts.
Requires an OpenAI API key, Go, and Docker.
It's recommended to run inside the sandbox.Dockerfile to prevent it from making changes to your workstation.