also, like what i want to do in the end, is somehow set up an ai shit like chatgpt, that basically lets me use a vision model when i upload an image (so it can see like a math problem or whatever), and agentic when it can do agentic shit, and use deepseek api so i can use a good model cuz i can't run beefy ones on my computer. any niggas experiment w this?
i KNOW theres a way to like, offload certain prompts to other models and have like, little baby models ask big models for help, and you need an mmproj so that the visionmodels can actually see images, etc. but yea no clue how you'd set that up or if there's even a way to.