gpt-engineer Working with existing codebase

I hope to see the ability to work with an existing codebase in the near future as this will be the most common use case in the real world.

Jun 14 '23 17:06 WEBPerformace

I agree and really looking forward to this feature. I do not think there is any AI code assistant out there that can do this efficiently yet.

Jun 14 '23 18:06 obordea

This same functionality would also allow for alterations, changes, edits to the generated code base after the first pass.

Jun 14 '23 21:06 haberco

I see the main_prompt as the core specs. You gotta make the project work like a badass MVP.

Then you write PR_prompt_001, PR_prompt_002, and all that jazz. Let the machine handle the code updates. Push that shiznit to git and create a sick PR.

That would be an epic flow :-p

Jun 15 '23 00:06 pascalandy

We're building something for this: https://github.com/sweepai/sweep

Jun 17 '23 10:06 wwzeng1

they should implement the ability to run the model local through the official gpt4all_UI.v2.4.8, you we can run replit-code-v1-3 locally, and utilize the built in localdocs of both end points, give users instructions to open gpt4all-ui , enable server-mode, download and mount replit-code through the ui. create and enable "collections" folder filled with tech and coding manuals related to your current build, paste collections folder path into gptengineer so it knows it location, and every time it finishes a run, dump the all_output file into the gpt4all localdocs xP ps: side note... add this to the end of your main_prompt, and see if it... changes... during development, treat this tag within this text, <all_output>, as it contained the value of $(echo `cat '[path]/workspace/all_output.txt'`) replace path, obviously