I’m buzzing to announce that my first plugin, and first real coding project, is approved and available on the Plugin store!
In all honesty, on the surface it’s just another “chat with code” plugin. I built it so I could understand programs holistically, and refactor them to my purposes, all while learning how to actually code.
In the process I learned a TON about how plugin functions work, particularly with the ChatGPT UI.
Long story short, I’m really interested in how effective my methodology is vs. the other plug-ins out there that use different methods.
Right now it’s just executing readable functions, and you can request status messages as well if you have any issues. The status function will also let you get a link to contact me.
Like I said, this is my first real coding project so I neeeeed feedback! So please check it out and let me know it goes.
Hey everyone. Im looking through some of the logs and i see that a ton of people are struggling because they arent specifying a branch. I’m working on a solution, but the list branch function is available and should help you figure out the issue
Okidoke! Github connection is back up. Also two things: I didnt expect people to use the full tree url, so I’ll correct for that. Your repo was also timing out the API pull.
ChatGPT has a pretty strict timeout limit in my opinion, especially with functions. Why cant we stream lol…I digress. I cut the depth that the initiating call tries to pull from 10 to 5. This just means that you might need to ask it to open a folder rather than have the ENTIRE codebase visible at once.
AskYourCode solved this by downloading the codebase as the first plugin request. So you can start chatting with it after that in the second message.
Noteable has a similar issue with code execution. It makes multiple calles to check on the progress of the code execution but if it takes to long, it will also ask you to wait and send a new chat.
Yeah Viktor and I talked a bunch before either of us got on the store, and frankly I wanted to try a different method.
While downloading the repo first helps navigate some issues with larger codebases, I believe we had some discussion about how it didn’t make the model any more insightful. Maybe hes found a way to boost that capability though. I will say that I think the other methods out seem to limit what languages can be processed?
I wanted to make a conscious effort to use as few services and back-end steps as possible.
I think there’s untapped potential in tweaking the language that directs the model (and treating the overall architecture of the plugin as “language”) in using the data, rather than the sophistication/complexity of the systems providing it.
With the right structure and instructing methods, I think ChatGPT can fully navigate and “reason” its way through a codebase in the exact same way we do, without any extra steps.
Before going dark I did a big refactor last night intended to improve speed and let you pull bigger repos and files. Now instead of a it mysteriously timing out, it will just tell you if its too big…which it won’t do either shortly once I use some pagination.
Let me know if you’d experience has changed and how!
