Fork of ChatGPT.nvim.
WARNING: this is a cut demo. In reality, it took almost 1.5 minutes to generate that code on my laptop.
gpt4all.nvim
is a Neovim plugin that allows you to interact with gpt4all language model. Unlike ChatGPT, gpt4all is FOSS and does not require remote servers.
-- Packer
use({
"WhiteBlackGoose/gpt4all.nvim",
config = function()
require("chatgpt").setup()
end,
requires = {
"MunifTanjim/nui.nvim",
"nvim-lua/plenary.nvim",
"nvim-telescope/telescope.nvim"
}
})
-- Lazy
{
"WhiteBlackGoose/gpt4all.nvim",
event = "VeryLazy",
config = function()
require("chatgpt").setup()
end,
dependencies = {
"MunifTanjim/nui.nvim",
"nvim-lua/plenary.nvim",
"nvim-telescope/telescope.nvim"
}
}
For configuration see upstream.
Plugin exposes following commands:
ChatGPT
command which opens interactive window using the gpt-3.5-turbo
model.
ChatGPTActAs
command which opens a prompt selection from Awesome ChatGPT Prompts to be used with the gpt-3.5-turbo
model.
ChatGPTEditWithInstructions
command which opens interactive window to edit selected text or whole window using the code-davinci-edit-002
model (GPT 3.5 fine-tuned for coding).
You can map it usig the Lua API, e.g. using which-key.nvim
:
local chatgpt = require("chatgpt")
wk.register({
p = {
name = "ChatGPT",
e = {
function()
chatgpt.edit_with_instructions()
end,
"Edit with instructions",
},
},
}, {
prefix = "<leader>",
mode = "v",
})
ChatGPTRun [action]
command which runs specific actions -- see actions.json
file for a detailed list. Available actions are:
grammar_correction
translate
keywords
docstring
add_tests
optimize_code
summarize
fix_bugs
explain_code
roxygen_edit
code_readability_analysis
-- see demo
All the above actions are using gpt-3.5-turbo
model.
It is possible to define custom actions with a JSON file. See actions.json
for an example. The path of custom actions can be set in the config (see actions_paths
field in the config example above).
An example of custom action may look like this: (#
marks comments)
{
"action_name": {
"type": "chat", # or "completion" or "edit"
"opts": {
"template": "A template using possible variable: {{filetype}} (neovim filetype), {{input}} (the selected text) an {{argument}} (provided on the command line)",
"strategy": "replace", # or "display" or "append" or "edit"
"params": { # parameters according to the official OpenAI API
"model": "gpt-3.5-turbo", # or any other model supported by `"type"` in the OpenAI API, use the playground for reference
"stop": [
"```" # a string used to stop the model
]
}
},
"args": {
"argument": {
"type": "strig",
"optional": "true",
"default": "some value"
}
}
}
}
The edit
strategy consists in showing the output side by side with the iput and
available for further editing requests.
For now, edit
strategy is implemented for chat
type only.
The display
strategy shows the output in a float window.
append
and replace
modify the text directly in the buffer.
When using GPT4ALL
and GPT4ALLEditWithInstructions
, the following
keybindings are available:
<C-Enter>
[Both] to submit.<C-y>
[Both] to copy/yank last answer.<C-o>
[Both] Toggle settings window.<Tab>
[Both] Cycle over windows.<C-m>
[Chat] Cycle over modes (center, stick to right).<C-c>
[Chat] to close chat window.<C-u>
[Chat] scroll up chat window.<C-d>
[Chat] scroll down chat window.<C-k>
[Chat] to copy/yank code from last answer.<C-n>
[Chat] Start new session.<C-d>
[Chat] draft message (create message without submitting it to server)<C-r>
[Chat] switch role (switch between user and assistant role to define a workflow)<C-s>
[Both] Toggle system message window.<C-i>
[Edit Window] use response as input.<C-d>
[Edit Window] view the diff between left and right panes and use diff-mode commands
When the setting window is opened (with <C-o>
), settigs can be modified by
pressing Enter
on the related config. Settings are saved across sections