👉 Support this work via GitHub Sponsors

A multi-llm Emacs comint shell, by me.

• ob-chatgpt-shell: Evaluate chatgpt-shell blocks as Emacs org babel blocks.
• ob-dall-e-shell: Evaluate DALL-E shell blocks as Emacs org babel blocks.
• dall-e-shell: An Emacs shell for OpenAI’s DALL-E.
• shell-maker: Create Emacs shells backed by either local or cloud services.

chatgpt-shell goes multi model 🎉

It’s been quite a bit of work, and I still have plenty of pending work. Please sponsor the project to make development + support sustainable.

My favourite model is missing.

chatgpt-shell is a comint shell. Bring your favourite Emacs shell flows along.

One shell to query all. Swap LLM provider and continue with your familiar flow.

chatgpt-shell includes a compose buffer experience. This is my favourite and most frequently used mechanism to interact with LLMs.

For example, select a region and invoke M-x chatgpt-shell-prompt-compose (C-c C-e is my preferred binding), and an editable buffer automatically copies the region and enables crafting a more thorough query. When ready, submit with the familiar C-c C-c binding. The buffer automatically becomes read-only and enables single-character bindings.

Navigate through source blocks (including previous submissions in history). Source blocks are automatically selected.

Reply with with follow-up requests using the r binding.

Want to ask for more of the same data? Press m to request more of it. This is handy to follow up on any kind of list (suggestion, candidates, results, etc).

I’m a big fan of quickly disposing of Emacs buffers with the q binding. chatgpt-shell compose buffers are no exception.

LLM being lazy and returning partial code? Press e to request entire snippet.

Request inline modifications, with explicit confirmation before accepting.

Both the shell and the compose buffers enable users to execute source blocks via C-c C-c, leveraging org babel.

I’ve been experimenting with image queries (currently ChatGPT only, please sponsor to help bring support for others).

Below is a handy integration to extract Japanese vocabulary. There’s also a generic image descriptor available via M-x chatgpt-shell-describe-image that works on any Emacs image (via dired, image buffer, point on image, or selecting a desktop region).

If you’re finding chatgpt-shell useful, help make the project sustainable and consider ✨sponsoring✨.

chatgpt-shell is in development. Please report issues or send pull requests for improvements.

Finding it useful? Like the package? I’d love to hear from you. Get in touch (Mastodon / Twitter / Bluesky / Reddit / Email).

• Load (require 'chatgpt-shell)

If using use-package, you can install with :ensure t.

(use-package chatgpt-shell
  :ensure t
  :custom
  ((chatgpt-shell-openai-key
    (lambda ()
      (auth-source-pass-get 'secret "openai-key")))))

You’ll first need to get a key from OpenAI.

;; if you are using the "pass" password manager
(setq chatgpt-shell-openai-key
      (lambda ()
        ;; (auth-source-pass-get 'secret "openai-key") ; alternative using pass support in auth-sources
        (nth 0 (process-lines "pass" "show" "openai-key"))))

;; or if using auth-sources, e.g., so the file ~/.authinfo has this line:
;;  machine api.openai.com password OPENAI_KEY
(setq chatgpt-shell-openai-key
      (auth-source-pick-first-password :host "api.openai.com"))

;; or same as previous but lazy loaded (prevents unexpected passphrase prompt)
(setq chatgpt-shell-openai-key
      (lambda ()
        (auth-source-pick-first-password :host "api.openai.com")))

M-x set-variable chatgpt-shell-openai-key

(setq chatgpt-shell-openai-key "my key")

(setq chatgpt-shell-openai-key (getenv "OPENAI_API_KEY"))

If you use ChatGPT through proxy service “https://api.chatgpt.domain.com”, set options like the following:

(use-package chatgpt-shell
  :ensure t
  :custom
  ((chatgpt-shell-api-url-base "https://api.chatgpt.domain.com")
   (chatgpt-shell-openai-key
    (lambda ()
      ;; Here the openai-key should be the proxy service key.
      (auth-source-pass-get 'secret "openai-key")))))

If your proxy service API path is not OpenAI ChatGPT default path like ”/v1/chat/completions”, then you can customize option chatgpt-shell-api-url-path.

Behind the scenes chatgpt-shell uses curl to send requests to the openai server. If you use ChatGPT through a HTTP proxy (for example you are in a corporate network and a HTTP proxy shields the corporate network from the internet), you need to tell curl to use the proxy via the curl option -x https://your_proxy. One way to do this is to set the proxy url via the customizable variable chatgpt-shell-additional-curl-options. If you set this variable via the Emacs Customize interface you should insert two separate items -x and https://your_proxy. See the curl manpage for more details and further options.

Endpoint: https://{your-resource-name}.openai.azure.com/openai/deployments/{deployment-id}/chat/completions?api-version={api-version}

Configure the following variables:

(setq chatgpt-shell-api-url-base "https://{your-resource-name}.openai.azure.com")
(setq chatgpt-shell-api-url-path "/openai/deployments/{deployment-id}/chat/completions?api-version={api-version}")
(setq chatgpt-shell-auth-header (lambda () (format "api-key: %s" (chatgpt-shell-openai-key))))

Launch with M-x chatgpt-shell.

Note: M-x chatgpt-shell keeps a single shell around, refocusing if needed. To launch multiple shells, use C-u M-x chatgpt-shell.

Type clear as a prompt.

ChatGPT> clear

Alternatively, use either M-x chatgpt-shell-clear-buffer or M-x comint-clear-buffer.

Save with M-x chatgpt-shell-save-session-transcript and restore with M-x chatgpt-shell-restore-session-from-transcript.

Some related values stored in shell-maker like shell-maker-transcript-default-path and shell-maker-forget-file-after-clear.

chatgpt-shell can either wait until the entire response is received before displaying, or it can progressively display as chunks arrive (streaming).

Streaming is enabled by default. (setq chatgpt-shell-streaming nil) to disable it.

There are more. Browse via M-x set-variable

If you’d prefer your own custom display function,

(setq chatgpt-shell-display-function #'my/chatgpt-shell-frame)

(defun my/chatgpt-shell-frame (bname)
  (let ((cur-f (selected-frame))
        (f (my/find-or-make-frame "chatgpt")))
    (select-frame-by-name "chatgpt")
    (pop-to-buffer-same-window bname)
    (set-frame-position f (/ (display-pixel-width) 2) 0)
    (set-frame-height f (frame-height cur-f))
    (set-frame-width f  (frame-width cur-f) 1)))

(defun my/find-or-make-frame (fname)
  (condition-case
      nil
      (select-frame-by-name fname)
    (error (make-frame `((name . ,fname))))))

Thanks to tuhdo for the custom display function.

Browse all available via M-x.

• Please go through this README to see if the feature is already supported.
• Need custom behaviour? Check out existing issues/feature requests. You may find solutions in discussions.

Pull requests are super welcome. Please reach out before getting started to make sure we’re not duplicating effort. Also search existing discussions.

Please share the entire snippet you’ve used to set chatgpt-shell up (but redact your key). Share any errors you encountered. Read on for sharing additional details.

Please enable M-x toggle-debug-on-error, reproduce the error, and share the stack trace.

Please enable logging (setq chatgpt-shell-logging t) and share the content of the *chatgpt-log* buffer in the bug report.

Please also share the entire org snippet.

👉 Find my work useful? Support this work via GitHub Sponsors or buy my iOS apps.

• Blog (xenodium.com)
• Blog (lmno.lol/alvaro)
• Plain Org (iOS)
• Flat Habits (iOS)
• Scratch (iOS)
• macosrec (macOS)
• Fresh Eyes (macOS)
• dwim-shell-command (Emacs)
• company-org-block (Emacs)
• org-block-capf (Emacs)
• ob-swiftui (Emacs)
• chatgpt-shell (Emacs)
• ready-player (Emacs)
• sqlite-mode-extras
• ob-chatgpt-shell (Emacs)
• dall-e-shell (Emacs)
• ob-dall-e-shell (Emacs)
• shell-maker (Emacs)

Made with contrib.rocks.