OpenAI releases new open-source agent orchestration library: Swarm

Posted by umarmnaq@reddit | LocalLLaMA | View on Reddit | 10 comments

[-]

medialoungeguy@reddit

They said they aren't reviewing PRs or issues.

This is just an attempt to find people that are curiously forking and making meaningful advances... then they will snatch the talent.

Smart move.

[-]

justletmefuckinggo@reddit

but does that also mean anyone can just have a new repo and collab off of swarm?

[-]

medialoungeguy@reddit

Sure, why not. Until it's soon to be outdated.

[-]

I have done something similar in my local setup. I am running 5 Ollama servers. First one is the primary generator, then there are summarizer, tagger, toxicity_checker, number_extractor, improvement_checker and so on running on multiple instances.

All of those are assistants, and I basically bounce requests from one to another.

Effects are great, just by using multiple assistants and extracting answers, tags and summary, I get fuller answers, but too many tokens. Also, I had to turn off streaming.

[-]