Hi,

I was using my search engine to look for available Emacs integrations for the open (and local) https://gpt4all.io/ when I realized that I could not find a single one.

Is there somebody who’s using GPT4All with Emacs already and did not publish his/her integration?

  • publicvoit@alien.topOPBEnglish
    1·
    1 year ago

    Hi,

    I personally would not have expected that the desktop app doesn’t have to run in background anyway. ;-)

    Any “gtp4all.el”-like mode would help me in writing my queries in Emacs as well as receiving its output directly into Emacs (babel/org-mode preferred, I suppose). Currently, I do a lot of copy&paste for that purpose.

    • karthink@alien.topBEnglish
      1·
      1 year ago

      In that case you can use it right now with gptel, which supports an Org interface for chat.

      Enable the server mode in the desktop app, and in Emacs, run

      (setq-default gptel-model "gpt4all-j-v1.3-groovy"
              gptel-host "http://localhost:4891/v1"
              gptel-api-key "--")

      Then you can spawn a dedicated chat buffer with M-x gptel or chat from any buffer by selecting a region of text and running M-x gptel-send.

      • ahyatt@alien.topBEnglish
        1·
        1 year ago

        It isn’t actually the same, though - they don’t support streaming. How are you getting around this?

        • karthink@alien.topBEnglish
          1·
          1 year ago

          In the meantime I added explicit support for GPT4All, the above instructions may be incorrect by the time you get to it. The Readme should have updated instructions (if it mentions support for local LLMs at all).

        • nickanderson5308@alien.topBEnglish
          1·
          1 year ago

          I have played with this a bit in the last few days.

          It’s nice and minimal, but I am hitting some issues with not enough memory. It seems gptel wants to load whatever model is specified, but I don’t have enough memory to run the model GPT4All desktop loads by default plus what gptel wants to load.