How do I assign backends & models to buffers programmatically?

[tomb7890]

Hi! I love, love, gptel! However there is one point of significant friction I have, which is assigning backends & models to Emacs buffers.

Below are two utility functions I've developed that set up the use of Gemini and ChatGpt.

(defun my/initialize-google-gemini ()
  "Set up gptel for use with Google Gemini."
  (interactive)
  (let ((api-key (password-store-get "google-gemini") ))
    (setq
     gptel-model 'gemini-1.5-pro-latest
     gptel-backend (gptel-make-gemini "Gemini"
                      :key api-key
                      :stream t))))
 


(defun my/initialize-chatgpt-4o-mini ()
  "Set up gptel for use with ChatGPT."
  (interactive)
  (setq gptel-api-key (password-store-get-field "openai" "apikey"))
  (setq gptel-model 'gpt-4o-mini)
)

I call one of them, and then do M-x gptel to set up a dedicated chat buffer for that specific backend & model. Then later I'll call the other defun, to prepare for the creation of a second buffer that uses the other model. I can then M-x gptel to create and use the 2nd back end and model. So far so good.

However if I now switch to the 1st gptel buffer, I find that now it too is using the 2nd backend & model.

I do know about the '=' feature in the transient menu, and I can use that to interactively fix up my confused buffers. But I've often not noticed this, which has led to frustration.

I would much prefer if my defuns could be modified so that this isn't necessary. Is it possible?

Thank you!

[karthink]

You are looking for setq-local.

(defun my/initialize-google-gemini ()
  "Set up gptel for use with Google Gemini."
  (interactive)
  (let ((api-key (password-store-get "google-gemini") ))
    (setq-local
     gptel-model 'gemini-1.5-pro-latest
     gptel-backend (gptel-make-gemini "Gemini"
                      :key api-key
                      :stream t))))
 
(defun my/initialize-chatgpt-4o-mini ()
  "Set up gptel for use with ChatGPT."
  (interactive)
  (setq-local gptel-api-key (password-store-get-field "openai" "apikey"))
  (setq-local gptel-model 'gpt-4o-mini))

Just remember to call this with the relevant buffer as the current one, for example via some hook.

It used to be that setting models/backends buffer-locally was the default. But that led many users to complain about the friction: to change the model across a project, they had to change the model manually in every project buffer. So I made it global and added the = switch. It was only a matter of time until someone found the global setting to be a point of friction 😄

But I've often not noticed this, which has led to frustration.

This wiki section may be of help.

[tomb7890]

Thank you so much Karthink!