I'm still fairly new to all of this, but your project jumped out at me, as it was very similar to what I had separately been starting to create. I've been running local LLM's a lot on my mac mini m4pro with 48G of memory, which, isn't bad, but I still run into context size limits frequently, and the agents just miss prior knowledge about things. Whats been helping is minimizing the number of AGENTS.md's in use, as well as making sure their sizes are just whats absolutely needed for context. Not sure of the possibility of showing in the logs and surfacing through the UI at some point, how much context was added for each query? as a means of maybe restructuring some of the pathways to minimize certain queries?
I'm still fairly new to all of this, but your project jumped out at me, as it was very similar to what I had separately been starting to create. I've been running local LLM's a lot on my mac mini m4pro with 48G of memory, which, isn't bad, but I still run into context size limits frequently, and the agents just miss prior knowledge about things. Whats been helping is minimizing the number of AGENTS.md's in use, as well as making sure their sizes are just whats absolutely needed for context. Not sure of the possibility of showing in the logs and surfacing through the UI at some point, how much context was added for each query? as a means of maybe restructuring some of the pathways to minimize certain queries?