test embbdeings models for replace llm for fast and low memory usage

start search for small vlm

make demo of the agent idea (using tools - python tools - extention tool - ocr and omniparser tools - small models )

try paddleocr