I’m working on a project of a hierarchical cooperative multi-agent framework that utilizes websockets for LLM<->LLM communication. Generally it is about running a websocket server that works as a ‘brain’ of the network and utilizes API endpoint(s) as a question-answering function for messages sent by multiple agents-clients connected to that server. Project is still in an early stage of development but I have already a working communication system - and it’s possibly the only currently available way to connect together completely different LLMs and let them speak with each other. Below is link to a folder in my repository where you can find couple different servers and clients (in Python, Node.js and html) most of which you can run as they are, since I try to avoid using OpenAI API and a simple schematics of the system as it is now:
NeuralGPT/Chat-center at main · CognitiveCodes/NeuralGPT (github.com)
And since I managed to get something what can be run and (more or less) works, I’m thinking about deploying the server in a HuggingFace space and (even better) get some kind of usable interface - and of course Gradio is the best way of doing it - especially that there are absolutely no issues with using a Gradio client as question-answering function for both: server and a client. Thing is that from what I learned Gradio interface uses websocket communication to have a working chat interface for LLMs - so there might be issues, is that correct?
Please help! I’m open to, any form of cooperation - it’s a quite ambitious and I’m just one guy who only just started working with code couple months ago… 