gpt.el is a simple Emacs package that lets you interact with instruction-following language models like GPT-4 and Claude 3.5 Sonnet from your editor. You can type a natural language command (with history and completion support) and optionally use the current region or all visible buffers as input for the model. The package displays the output of the model in a temporary buffer, and updates it as the model generates more text. You can issue follow-up commands that provide the interaction history in that buffer as context. You can also browse, save, and clear the command history for later reference.
To use gpt.el, you need to have at least one of the openai
and anthropic
Python packges as well as jsonlines
. You'll also need valid API keys for OpenAI or Anthropic. You can install the packages with:
pip install openai anthropic jsonlines
You can get an OpenAI API key from https://beta.openai.com/ and an Anthropic API key from https://console.anthropic.com.
MELPA is a popular third-party package repository for Emacs. To install gpt.el from MELPA, first add MELPA as a source in your Emacs init file:
(require 'package)
(add-to-list 'package-archives
'("melpa" . "https://melpa.org/packages/") t)
Then, use the built-in package manager to install gpt.el:
M-x package-install RET gpt RET
Once installed, you can require the package in your init file:
(require 'gpt)
Alternatively, you can use use-package
:
(use-package gpt)
To install gpt.el, clone this repository and add the following to your Emacs init file:
(add-to-list 'load-path "/path/to/gpt.el")
(require 'gpt)
Alternatively, you can use use-package
:
(use-package gpt
:load-path "/path/to/gpt.el")
You need to set the variables gpt-openai-key
and/or gpt-anthropic-key
to your respective API keys to use gpt.el. For example:
(setq gpt-openai-key "sk-Aes.....AV8qzL")
(setq gpt-anthropic-key "sk-ant-api03-...")
Optionally, you can customize the model parameters by setting the variables gpt-model
, gpt-max-tokens
, and gpt-temperature
. The defaults are:
(setq gpt-model "gpt-4o")
(setq gpt-max-tokens 2000)
(setq gpt-temperature 0)
By default, gpt.el uses the OpenAI API. To switch to the Anthropic API, you can set:
(setq gpt-api-type 'anthropic)
To run a generative model command, use the gpt-dwim
function. You can bind it to a key of your choice, for example:
(global-set-key (kbd "M-C-g") 'gpt-dwim)
When you invoke gpt-dwim
, you will be prompted for a command, with history and completion. The command can be any text. For example:
Write a haiku about Emacs.
If you have an active region, it will be used as contextual input to the command. If you enter n/a as the command, only the region will be passed to the model. The output of the model running the command will be displayed in a temporary buffer, with the same major mode as the original buffer. The output will be streamed as it is produced by the generative model. You can switch back to the original buffer at any time.
If you want to use all visible buffers as input, use the gpt-dwim-all-buffers
function. You can bind it to a key of your choice, for example:
(global-set-key (kbd "M-C-b") 'gpt-dwim-all-buffers)
In the gpt-output buffer, C-c C-c
is bound to running a follow-up command that is provided the previous commands and outputs as input. For example, you can run a command "Explain this in more detail" to get more information about the previous response.
In the gpt-output buffer, C-c C-b
is bound to copying the content of the code block at point to the clipboard.
In the gpt-output buffer, C-c C-p
is bound to toggling the visibility of the "User:", "Human:", and "Assistant:" prefixes.
You can switch between different models (GPT-4, GPT-3.5-turbo, and Claude 3 Sonnet) using the gpt-switch-model
function. In the gpt-output buffer, C-c C-m
is bound to this function. When called, it will prompt you to choose a model from the available options.
You can view the command history by calling gpt-display-command-history
, which will show the commands in a buffer. You can also export the command history to a file by calling gpt-export-history
, which will prompt you for a file name. To clear the command history, use the gpt-clear-command-history
function.
Contributions to gpt.el are welcome! Please feel free to submit pull requests or create issues for bugs and feature requests.
gpt.el is licensed under the MIT License. See LICENSE.md for details.