Hey Alex, I love the concept myself! I can share a few personal learnings while building my own project.
Combining conversational and drag-and-drop experiences is a good and well-accepted balance by users as it provides a "template" like starting point that you are able to customize further at will.
AI-based section generation for the site is huge but if it is too open-ended users shy away. It would work better by having users provide context in a non-conversational way. For example instead of having to prompt the user so they type "can you please create a section of x with layout of y and colors of z" have them pre-select the option from a dropdown "new section" and describe it. Or a "new template" and ask for an inspiration e.g "Inspired in space x with a pricing section x,y,z". Conversation is powerful to describe ambiguous things but people will get frustrated chatting through an experience that can be more handrailed.
Look into how to use AI assistance to handle other tedious parts of the GrapesJS experience for example, CSS classes. You can have your users visually describe the appearance and paste those styles into existing elements in the canvas e.g: add a gradient of 4 blue colors to the selected component. Or control themes.
I love the direction you are taking the project and forget about the saturation. Website builders is a good niche because there is a huge demand. You just need to find the right starting audience, get quality feedback and keep iterating.
Careful marrying too much to GrapesJS.
