Yesterday I got a bit distracted with videos on Ai agents, then had a quick gander at some of the options. One of the main things I found out is that the industry is pretty much being choked by GPT-4.
Now I guess these things do not come cheap, but if you look at https://huggingface.co/ there are plenty of free models and even free APIs to speak to them. Now sure they are not GPT-4 level but some of the specific ones from what I gather can do what they need to do without too much fuss.
I guess the reliance and almost exclusive support for OpenAi GPT is on one hand a quality of data thing, and on the other, it is a recurrence of the shit mentality behind EVM and Ethereum domination for chads.
With that said, if you have an OpenAI API key and pay them, then no biggy. For others, you can probably get a new one with $18 credit to test out things more hands on. If you already have an account and never knew you got credits, you will need a second phone number.
Ok... I gave up, and got some credits because frankly, I do not have the hardware to deal with local models.
LM Studio
If you do have the hardware then LM Studio seems to be a great local way to play with models.
This is exactly the type of backend that will eventually drive multi-disciplinary AI workers. LM Studio exposes a GPT compatible API which you could have very specific models be accessible on and easily iterate over testing newer and better versions.
On top of that is where AI Agents come in, and although most of the next tools mainly support GPT-4 I think a push for more interoperable AI API usage and standards will be adopted. Ie the same EVM shit we have with Ethereum.
SuperAgi
SuperAgi is one of the only ones I found really trying to build something open and accessible, they have recently push their cloud interface and although it will surely cost a penny in the future, seems to be pretty much free for now.
This allows a person to jump in rather quickly and play with the concept of AI Agents. The only caveat which I bitch about more than enough already is the strong focus on GPT support.
I could not get the cloud one to work using HuggingFace although it does say it supports it. Not a biggy though I am sure if I were to instead try to run their system local I could poke around more and maybe it will work with only the HuggingFace.
For now I just ran one of their default agent templates, the ContentIdeaGenerator which you can find in the market place as a test, and likely expand on it with bits to see how these Agents truly work together.
WTF is an Agent?
Oh? I did not say?
Well, think of them as the slaves you instruct to do things in shops every day and then say things like "I pay your salary". Just without the last part.
It is a method of orchestrating a goal-orientated workflow between many individual AI bots that perform their individual tasks and iterate until the final goal has been satisfied.
Then your next question is: "Isn't this just the logical way to go about things, why are you covering old news?"
Suck it.
These are new tools, built to manage multiple Agents in a user-friendly manner. So yeah old news but new tools.
Now where was I...
Rivet
Rivet is a visual node-based approach to doing the same thing. I think SuperAgi does a good job of compartmentalizing each aspect of your setup, but a node-based approach might help visualize more complex expectations.
Rivet does run locally and currently strictly speaks to OpenAI although there seem to be a few workarounds to make it speak to other Api's.
FlowiseAi
Flowise also has a node implementation but the difference with them is that they run as a basic node application. They do seem to rely on a few more external tools and have a far more involved setup but it is good to mention them if you want to give it a go.
ChatDev
ChatDev is a fun one, and with all this, I have either just watched vids or poked around in the docs but I think ChatDev is the most likely to be something you use to produce actual tools.
The others are good at performing tasks but ChatDev is specifically designed to build software and more hands-on products.
You are essentially orchestrating an office environment with each bot having a speciality, the difference I see from the others is that these produce actual working software. As in you can click an exe and it will run it on your PC.
In their own words:
So that is it for a list of pretty user-friendly AI tools to accomplish more than asking ChatGPT how rich Elon is. I personally have not cared for all these read my emails or create Tweet-type things, but with these Agents becoming easier to orchestrate a person can without a doubt run every aspect of whatever grinding tasks can think of.
Goodluck.
All images are screenshots