How to colorize `gptel` response persistent

[tshu-w]

I'm using gptel to refine my writing. Sometimes I modify and resend responses to OpenAI to get new replies. I want to add a face to the response text so that I can realize clear text properties when resending the modified response. Otherwise, gptel will divide it into multiple prompts. I tried the following code but it didn't work. Any suggestions would be greatly appreciated.

  (add-to-list 'gptel-response-filter-functions
               (defun gptel--colorize-response (content buffer)
                 (put-text-property 0 (length content) 'face 'font-lock-warning-face content)
                 content))

[karthink]

If I understand correctly, you want to classify a response message from ChatGPT as a user prompt.

For the purpose of (repeatedly) rewriting text, you could just use the rewrite feature built into gptel. Select the text to be rewritten, then gptel-menu -> r -> R. This should work no matter the source of the text selected (i.e. ChatGPT or you). Is there some reason you don't want to do it this way?

[tshu-w]

If I understand correctly, you want to classify a response message from ChatGPT as a user prompt.

Yes, this is what I need, or gptel consider all content as 'user' with an option, regardless of its property.

I am not using the default rewrite because I have many different directives that implement "rewriting", such as translation, paraphrasing, abbreviation, etc.

[PalaceChan]

(maybe misunderstanding but if you universal prefix arg before the rewrite can't you edit the prompt to select on of these many rewrite directives?)

[tshu-w]

maybe misunderstanding but if you universal prefix arg before the rewrite can't you edit the prompt to select on of these many rewrite directives?

I didn't realize before that the rewrite directive can also be modified, but it's not possible to select "gptel-directives" and must be manually entered, which makes it difficult to switch directives frequently.

[karthink]

Yes, the UI for selecting rewrite directives needs to be better. I'll try to improve it soon.

On Mon, Jun 12, 2023, 6:55 AM Tianshu Wang @.***> wrote:

maybe misunderstanding but if you universal prefix arg before the rewrite can't you edit the prompt to select on of these many rewrite directives?

I didn't realize before that the rewrite directive can also be modified, but it's not possible to select "gptel-directives" and must be manually entered, which makes it difficult to switch directives frequently. [image: Screenshot 2023-06-12 at 21 52 47] https://user-images.githubusercontent.com/13161779/245165265-d763f69c-fd19-4321-8807-3b7195b85a82.png

— Reply to this email directly, view it on GitHub https://github.com/karthink/gptel/issues/73#issuecomment-1587390655, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACBVOLBRK7NGVIJUIU5JAXDXK4NUHANCNFSM6AAAAAAZC3VJCU . You are receiving this because you commented.Message ID: @.***>

[karthink]

but it's not possible to select "gptel-directives" and must be manually entered, which makes it difficult to switch directives frequently.

I took a look at this, and realized the history of rewrite directives is already saved. You can access all previously used directives with M-n and M-p in the minibuffer prompt. If you use savehist-mode, it is persisted across Emacs sessions.

[tshu-w]

Putting aside the original intention, I still want to know if there is a way to set up the face of the reply, so as to better distinguish between the original content and the reply.

[karthink]

You could try something like this:

(defun gptel--color-response ()
  (let ((prop))
    (save-excursion
      (setq prop (text-property-search-forward 'gptel 'response t))
      (put-text-property (prop-match-beginning prop)
                         (prop-match-end prop)
                         'font-lock-face
                         'warning))))

(add-hook 'gptel-post-response-hook #'gptel--color-response t)

Replace the warning face with whatever you need.