Intro and a few questions

[Tokoyuma]

Hi all,

I'm new to this project and am very interested in it and the possibilities it opens up. For background, I'm a solutions architect, familiar with many stacks, platforms, APIS, and the software development lifecycle. I'm not an expert coder but I can hold my own and I have a knack for understanding complex problems and coming up with possible routes to move the solution foward.

I don't have gpt+ (and they've stopped accepting upgrades temporarily) so I'm reading the repo by hand instead of using the various assistants. As soon as they allow upgrades to gpt+, I'll be able to come up to speed faster.

I'm starting to understand the system arch of the swarm (SOB,Execs,Agents) and their reasons for existing. But I want to better understand the relationship between what a Swarm is and how(if) it is different than an Agent. Is it possible that any given Agent can itself be built and executed as it's own swarm? Maybe it's semantics but I want to make sure I'm not making it more complex than it needs to be (in my head).

I'm also trying to understand the distinction between code in the repo, and the use of the GPTs (or other models if that's the case). How is the code used vs the usage of actual GPTs/agent/assistants/etc? Does the code need to installed and exist on local hardware? I'm sure I'll figure this out as I review the code files that have been committed so far.

It's possible these have already been answered or discussed and I just haven't found it yet.

Anyway, wanted to stop by and introduce myself.

Cheers,
Bob

[heretolearnaboutLLMapps]

Hey Bob,

Trying to answering only your question about swarm and agent definitions:

1. A swarm is an organization of multiple agents.
2. A swarm of swarms would be a super organization with each individual organization would be akin to an agent, but we're not calling it that.
3. Agent would be the most basic unit, or the individual 'worker'

[FireMMDC]

An agent is an entity that has the ability to take actions within an Environment.

This project is using gpt-4/gpt-3.5 as the reasoning for an agent within the swarm via the Assistant API https://platform.openai.com/docs/assistants/overview.
The Assistant API is basically a structured way of giving the gpt models a persona, as well as providing them with a list of resources they are able to access, either text documents or function calls they can retrieve information from that are present within the calling script.

A single assistant has the ability to be used in multiple context windows at the same time via the Threads construct https://platform.openai.com/docs/assistants/how-it-works/managing-threads-and-messages.

The code in the repo contains experiments for ways to set up communication between multiple gpt instances.
It has an example of a Tool Maker/Function Builder pair of Assistants.
It also has a script to create to Assistants within your own OpenAI account based on the definitions for some of the Agents expected to be within the HAAS.

To work with some of the scripts I think you just need to get credits with OpenAI, https://platform.openai.com/playground , I'm not sure if GPT+ provides API credits.