This allows for easy communication between emacs and the openAI API platform; allows using all of the available models and integrates cleanly with emacs toolings. Here is the basic things you will need for set-up:
- Install the package from MELPA using:
(package-install 'gptai)
then put this in your init.el:
(require 'gptai) ;; set configurations (setq gptai-model "<MODEL-HERE>") (setq gptai-username "<USERNAME-HERE>") (setq gptai-api-key "<API-KEY-HERE>") ;; set keybindings optionally (global-set-key (kbd "C-c o") 'gptai-send-query)
To fill out the details of the configuration section
- Define the desired model to use; available models can be found by running
gptai-list-models, it will also display this list in the gptai buffer. Use this to choose a model and setgptai-model(text-davinci-003 is a good default model). - Define your OpenAI username.
- Define your API key (your own key can be obtained from OpenAI API Keys)
After that optionally fill out the init section with some keybindings.
Check out the wiki for examples on how to easily extend this package to your uses
You can send textual queries to different models of openAI using the functions:
gptai-send-queryPrompts in the minibuffer a query and returns the output at the point.
gptai-send-query-from-selectionSends the text in the current selection to openAI as the prompt to the openAI API model you specified in your configurations, returns the output in place of the original selection.
gptai-send-query-from-bufferSends the current buffer as the prompt to the openAI API model you specified in your configurations, returns the output in place of the original buffer.
gptai-spellcheck-text-from-selectionSends the selected text for spellchecking to openAI and replaces the selection with the corrected text.
gptai-elaborate-on-text-from-selectionSends the selected text for further elaboration by openAI and replaces the original selection with the improved text.
You can generate images with the DALL-E engine using:
gptai-send-image-queryPrompts the user in the minibuffer for
- A prompt for the command
- How many images you want to generate
- What size they should be
- Where you want to store them on the disk
It will then generate those images at the API endpoint, then use curl to download those images to your specified directory path. Once it is done, if one image was downloaded it will open it in a new buffer for viewing, otherwise if more were downloaded it will simply display success when done.
Using the gptai-code-query and gptai-code-query-from-selection functions,
you are able to generate code from instructive prompts in specified languages,
both of these functions by default will prompt the user for a language and
return the returned code in-place at your selection or point in the buffer.
gptai-code-queryPrompts the user for instructions and language to use; then returns the output code that was generated at the point.
gptai-code-query-from-selectionUses the selection text as the instructions and prompts for a language to use; then returns the code that was generated in place of the original selection.
gptai-explain-code-from-selectionExplain the code from the selection, return output above selection.
gptai-fix-code-from-selectionFix and debug the code from the selection, return the output in place of the original selection.
gptai-document-code-from-selectionDocument and describe the code from the selection, return output above selection.
gptai-optimize-code-from-selectionOptimizes and refactors code from selection, returns output in place of the original selection.
gptai-improve-code-from-selectionImproves and extends on code from selection, returns output in place of original selection.
The newer models of gpt-3.5 and gpt-4 requires a different query structure, this
means that to use these models there needed to be a seperate query handler; this
is achieved with gpt-turbo.el. Using the turbo model is as easy as calling
gptai-turbo-query while passing it a text query input either interactively or
pragrammatically. Here is an example:
(gptai-turbo-query "This is a prompt")this functions basically the same as gptai-send-query, it just limits you to the newer 3.5 turbo, and 4 models for GPT language models instead of giving more ability to choose what models you are using with the standard query handler.
Feel free to make a PR with improvements, all PRs should include your changes as well as a addition to the CHANGELOG.md file noting any important changes for users to be aware of.