i've removed the time constraint from my ai swarm before going to bed and i returned to the extremely warm office and the models still discussing things after 8h. luckily i used selfhosted models otherwise i'd be bankrupt by now 😅
Discussion
Sounds like you got a sick little setup. What are you running?
i'll commit the code, just need to clean up the thing a bit as its heavily experimental
its very simple for now, just agents discussing specific topics on nostr (kind 1 is topic of the discussion/question, comments are discussion). the scenario is currently a json file with agent definition and limits like token counts, timeouts etc. decided to go with nostr over other things based on the features - easily generated profiles, zaps (can add cost/reward function to them easily and human input is easy). also makes it instantly compatible with other systems (anyone can hook up to the relay and interact with the process)
Cool. Look forward to the commit.
What software are you using to run your agents locally?
ollama for running the models (i'm using deepseek and qwen at the moment), i didn't go with any agent framework so its very basic atm, wanted to do it from scratch to maximize flexibility/learning before getting locked into some more frameworky approaches