Non-modal interface

[mityax]

Describe the need of your request

I often use codegpt to generate documentation comments. However, the current modal interface requires more clicks if I can't afford to keep it open all the time (which I can't, because it occupies so much screen space), as I have to close (and potentially also clear the chat) when I'm done.

Proposed solution

I'd really like a non-modal code generation interface. As an example, take the notion AI interface:

https://github.com/user-attachments/assets/bb738678-3051-4dd1-81b1-30c2e9d966fb

When clicking "Insert below" here, however, only the generated code should be inserted, not the entire reponse. Addionally, an "insert above" action would be very useful as well (e.g. when generating a documentation comment).

Additional context

Two more additions that would play nicely with a non-modal interface:

• I'd very much like the CodeGPT editor actions to have a field on whether they need context (or even better: how much context). If you could choose for each editor action whether it needs the surrounding methods or the entire surrounding file as context, that'd allow generating code/documentation without having to select the code required as context, thus the cursor could stay at the right insertion point.
• A shortcut for opening the non-modal interface (instead of having to right-click like you afaik have to at the moment to use Editor Actions). I'd suggest Alt + Space by default, as that'd feel consistent with the Alt + Enter shortcut intellij IDEs already have.

[carlrobertoh]

This seems like something the Edit Code feature would handle well, but it is currently only supported for the CodeGPT provider and needs a bit of a touch, as you can see in the video:

https://github.com/user-attachments/assets/437f31d7-47df-4cd3-82bb-3edf8225d5f4

However, I do like the idea of quick inserts and I believe the same UI/UX concept can be used for your described feature as well.

[mityax]

Looks good, I haven't used this yet since I use the ollama provider.

As a quick note, I think it'd be great if there was a way to insert code without rewriting existing code, as this might lead to bugs being introduced without one noticing, since there's always a risk that the llm dies not exactly reproduce the existing code