This is an interesting analysis. Many people rush to use paid APIs without exploring local options, missing out on understanding model behavior. Running smaller models in LM Studio or GPT4All is a great first step for developing agents.
I also agree about the importance of maintaining context. Not storing sessions or fine-tuning on your own data can lead to losing important information. A lightweight model can serve as a memory layer while a larger model handles reasoning.
Thanks for sharing this perspective; it’s crucial to try local models first.