Would it double your codebase? Do you think it would work for a large codebase?
Not anywhere close. It's basically just maintaining a simple descriptive index the model can later use to decide what files it needs to read given the task you've given it.