Add window splits to context

[svermeulen]

I often am porting code from one file to another, which is a perfect application for AI. I find it necessary in these cases to copy and paste the file I'm porting into the new file as a big comment, so that the AI can get this information in its context. Would be nice if this was automatic, since it's very common that any window splits in the current tab are relevant.

[milanglacier]

This is a useful feature, and I agree. I’ll find a time to develop it, also make it as out-of-the-box as possible. But there is no specific timeline for this. PRs for the implementation are also welcome, of course.

However, please note that depending on your chosen model—particularly with Codestral or local LLMs like Ollama or llama.cpp—you may still need to write some boilerplate setup code.

For context, Minuet supports two types of LLM requests:

1. Chat/Instruction Requests: The most common type for general AI development, where you provide instructions for the model to follow.
2. FIM (Fill-in-the-Middle) Requests: Primarily used for code completion and less common in other AI workflows.

For a detailed comparison between the two requests, refer to this discussion.

With chat/instruction requests, we can include a built-in method to generate prompts incorporating content from other buffers , as we need to provide instructions to the chat-LLM directly.

In contrast, FIM requests demand strict adherence to model-specific special tokens and formats to ensure the generated text conforms to the required structure. Since these formats vary across different models, there is no universal method for constructing the necessary input. Once this feature is implemented, I plan to provide an example configuration that incorporates context from other buffers, which users can then adapt according to the specific models they are using.