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.
Thanks for the feedback! I want to clarify: HA and SSO aren't implemented yet, and I haven't finalized the pricing model. The "enterprise pricing" mention was exploratory—I'm still figuring out sustainable revenue options.
My goal is to keep everything open source if possible, including HA and SSO when they're built. I'm a solo dev trying to balance sustainability with accessibility for SMBs. If you have ideas on how to achieve this, I'd genuinely appreciate an email.
I'll make mistakes along the way, but I'm committed to listening and improving.
Deve here again. Interesting, I never heard about KASM Workspaces. I can not give you a good comparison since I did not use it. But based on what I understand, yes, we are similars in the Desktop as a Service. In my case, I'm aiming to VDI, not DAAS, but is kind of the same.
The main difference is the target user. I'm aiming for a non-enterprise user, a more normal PC user. I want to make the UI and lifecicle as simple as possible. I don't think in today state I reached what I wanted, but not a bad starting point. Users will tell me.
The dev here. Thaks for the feedback. I'll try to do better next time. I'm a single dev with cero knowledge of video editing (as you clearly noticed :P).
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