Pass full local files to OpenAI API

[rambalachandran]

Describe the feature

One of the important limitations that I have in the current implementation is that I would like to pass json/csv files of input data and expected output data to OpenAI API and ask it to generate a function or method to perform the transformation. VSCode has access to all files in the current workspacefolder and being able to pass them to OpenAI API would be very helpful

[Christopher-Hayes]

You can already do this with "Editor selection".

But, I'll leave this issue open for an easier way to select (multiple) files. The thing is adding entire files is a quick way to max out token count. But, the nice thing about CSVs is that you can just give it the header row, and maybe a couple real data rows, and usually that's all it needs, so providing the whole data file is often not needed.

[Christopher-Hayes]

I understand this ask is partly coming from expectations that ChatGPT Reborn do a lot of the same things that the ChatGPT Web UI can do, so I'll look for ways to make the UI feel more familiar to ChatGPT Web UI users.

[raphael2692]

Hi @Christopher-Hayes , I read that chromadb supports a Javascript Client. I am not experienced with vscode exstensions, do you think we could integrate it in ChatGPT Reborn? The usecase would be: on button press embedd a set of (supported) local files and the open a chat panel to chat with them.

My only concern is that iusing a local model for embedding is tassative on cpu while using remote services like chatgpt is expensive...

https://github.com/chroma-core/chroma

[Christopher-Hayes]

@raphael2692 yeah looks possible. The db seems like the easier part, the harder part is knowing when to use text embeddings and making work in a way that isn't confusing for the end user.

[wojtekcz]

The thing is adding entire files is a quick way to max out token count.

One can use Claude with 100k context window. That would push token limit specter back a bit ;-)

[Christopher-Hayes]

Marking planned, but this extension needs to support the "Assistants API" first. So, that needs to come first, once we support it, uploading files to the assistant should easy.