Seems like we should prioritize working on the Templates that improve the output as the system prompt. I am using cursor.directly which helped a tiny bit but still can get even remotely as close to bolt.new outputs even by providing additional prompting that bolt.new adds to response message.
What am I doing wrong here
Can you explain / show bit more what you trying to achieve?
Starter Templates are already there and can be used:
Can you share the prompt as well?
And did you also use the same Model as bolt.new uses (Anthropic Claude Sonnet 3.5 (old))?
There are other things which lead to different outputs, but at least the testet model should be the same. Otherwise you cant expect the same at all.
Same model and exact same prompts.
Are there libraries or templates that I need to pre-install to get better more hi fidelity outputs similar to Bolt.New?
I far as I know we dont use the same templates, but I am not sure. Maybe @thecodacus or @aliasfox can answer this.
Other then that, you could just create a new project in bolt.new by clicking their template and then export it without prompting anything.
Import it in bolt.diy and you have the same base
basically the difference is that bolt.new does not start from scratch, it imports a template and then start editing that template to get a head start.
we have added this feature as an optional non default option (still experimental).
try using the starter templates to build the application
since it uses github api, for better increased rate limit, I recommend setting up github access token from connections for this
@thecodacus thanks, but the quesiton was more, if we use 100% same starter templates as bolt.new has, or do we have other templates, we pull from git 
we have some of the same starter templates, excluding the nextjs one. as it has shadcn/ui components preloaded in the template, not all the models can load them all at once in context, so this was currently excluded in the auto load. But that template gives the best results.
Thanks for the explanation. Could we load shadcn/ui components preload at least for Ollama models soon?
@itqandas just create an feature request / discussion here: [bolt.diy] New Features Discussion - oTTomator Community
then we can discuss it there and maybe consider for the planning of the roadmap / next features.
