Not directly. Save or export the page as an image (PNG or JPG, around 3000 pixels on the long side) and load that. If the PDF is a scaled drawing, set the scale with two clicks on a dimension you know.

On clean plans most walls, doors and windows are found automatically. Busy, small or blurry plans need more fixing. That is why there is the Check & Fix step: you see the result on the plan and correct it with a few clicks before building.

No. Everything runs inside Blender on your own computer. Recognition is ordinary image analysis written in Python, there's no AI model and no online service behind it.