I've being building a personal proyect for a while already, with the objective of having my own team of researchers and developers running entirely on small LLMS (14b or so params).

I thoug to keep it private, but I'm not planing to make money with this, so... why not making it public?

In the repo, there is a file called `architecture.md` where I explain the main points of how does it works, the main idea of the system and why is a big different to any other agen swarm over there.

But the first point is that I'm aiming 100% to small models. That does not means that this project does not work with gemini, openai or anthropic, it means that I don't use them as a comparison to say that something is solve or not. Maybe it works with gemini, but not with lets say qwen 3.5 14b. I mean, yes, qwen 3.5 is really good, but we can not compare a hudreds billon params or even trillon params to a <100b params.

The way of work and prompting to small models and particularly, agents build on top of small models is substancially different.

The project is just a fun project to learn about swarm, prompting and small LLMs.

I hope you enjoy it as I enjoy it working on it.

If you have any question, just write it down.

https://github.com/Infinibay/researcher