Experimenting with SuperAGI

We are planning to build a new tool for bolt.fun that is AI based, & we wanted to see if SuperAGI can help us with that. So that's why I've been today playing around, & reading about its different capabilities to see if it's the right tool or not.

What is SuperAGI??

It's a framework that enables you to structure & deploy autonomous AI agents quickly & these agents have access to some cool capabilities.

Its Best Features:

- It can access & use external APIs by itself.

Not any external tool of course, there are a pre-determined set of tools like: Read/Write file, Google Search, Send Email, Send Tweet,...etc.

A list of the tools SuperAGI provides

Of course, you need to give it access to your google account, twitter account, ...etc, if you want it to be able to automatically post tweets, send emails,...etc.

- It's easy to use (I'm talking about the GUI specifically)

There are not a lot of configuration you need to do to get it to start working.

How it works

Whenever you run the agent, you need to provide it with some inputs in some form (e.g. a txt file).

After that, the agent will run in MULTIPLE steps to achieve its goals.

In each step (iteration), it either calls the LLM, or calls some of the external tools.

It's up to the agent to decide what is best to do next, or what tool to use.

Then it feeds the results of the previous steps (selectively), into the next iteration.

& this keeps going on until either the agent determines that it reach its goals, or it reach the maximum number of iterations it's allowed to perform.

It's needless to say that executing a run can consume a lot of tokens from your API key, & this number increase the more iterations the model is allowed to run.

What I had to do to get an agent up and running?

- After signing in using your Github account, you are taken to a dashboard where you can click on 'Create agent'.

- You can either start with a predefined template for the agent, or build one from scratch.

A list of templates that you can use to build your agent on

- After that, you are asked to fill a few fields to tell this agent how it's supposed to work.

Fields like:

Agent Name
Description
List of goals
list of instructions
What AI Model to use
What external tools it has access to
List of constraints

The fields you need to fill to create a new agent.

- Finally, you just hit "Create and Run"

- The agent will take the input, & start iterating, until it reaches its goals.

Using as Code

I've been primarily testing the GUI, but in fact, SuperAGI is a code framework first & foremost, & its GUI has only been released recently.

So there might be more options & features that you can do in code which you can't currently do from the GUI.