Right now I get weird responses like when I tell it to analyze a file or a set of files, it tells me it does not have access to the file.
boltdiy is not fully feature rich like the bolt product from stackblitz. the editor and agent still don’t talk to each other that way. not yet.
I get the same error in bolt.new, where asking the the AI to review a part of the codebase simply does not work. That seems preposterous for an AI integrated IDE.
The enduser/poweruser needs to have the full ability to direct the AI to digest the contents of the webcontainer (and therefore review any file in the codebase and create responses based on the directed review of codebase files).
I understand there is a context window limitation but that can be worked around by the poweruser with an folder root/.ignore file.
enjoy this emoji
Proposal for
Settings → Features → Append WebContainer (Package) To (The End Of) Prompts
Allow me to provide a greater overview of the situation and why this feature is so important
When I want to use the artificial intelligence for advanced research projects or for extremely sophisticated segments of a code base I need to be able to completely guide the artificial intelligence With a set Of specific instructions and encyclopedic knowledge in the form of those specific instructions so that the artificial intelligence can properly do the micro job that I have assigned to it in any given, particular prompt.
I believe it is one of the greatest virtues of this free ad open software project is that such a thing as bolt.diy deployments is even comprehendible.
Therefore I believe this is a critical feature such a feature that allows the end user to toggle a functionality that cuts out all thinking phases and prompt enhancement features of the local bolt.diy client in order to ensure the complete correct delivery of the project context in this case being the complete WebContainer contents as project context.
Also, ought all of the file names, contents, along with their absolute directory pointers of the web container project files be delivered to the AI to ensure easy of reference between enduser and AI.
If this helps provide any color I think of the AI as existing in its web container and therefore I conceive of the AI as being able to take the unique perspective utilizing all of its intelligence from its greater compute power and from its training at origin, and,
Focusing all of that to a laser point inside of the code base without losing the world of the code base that is what I mean by leveraging the power of the AI.
The specific instructions and files and information such as rolodex information or such a specific information about how the project code bases factory Singleton and other abstract patterns are implemented is critically important and often lost in the non-simplified Project Files consumption approach, of which I am proposing the toggle for.
That is what I mean by the importance of a feature that permits the end user To be sure that the AI is consuming the entire web container code base contents and files.
With a toggle option to use this simple approach verses all of the wonderful features such as the thinking processes and token saving context features that while effective for vibe coding and general code creation by the AI are not compatible with the true full access to the web container and its files with each prompt approach.
Wherein using the toggle to send the webcontainer contents package with every prompt it can be more or less assured the AI will respect the embedded programming patterns in the project files.
Whether or not Settings → Features → Context Optimization accomplishes this or not is unclear. A setting name for the feature I am putting forward is more along the lines of Settings → Features → Append WebContainer (Package) To (The End Of) Prompts.
The input to the AI must have a reference name that may be used in prompts such as “Focus entirely on the Project Files Package (PF Package; PFP).”
submit pull request when you’re done
yapping**************
as far as I know this is related to the context optimization feature in Settings. If you turn this off, it will use everything to send to the LLM.
Did you try this out?