vim ai
1.0.0
Ce plugin ajoute des capacités d'intelligence artificielle (AI) à votre VIM et Neovim. Vous pouvez générer du code, modifier du texte ou avoir une conversation interactive avec les modèles GPT, tous alimentés par l'API d'Openai.
Pour avoir une idée de ce qui est possible à faire avec les commandes d'IA, voir les invites sur le wiki communautaire
Ce plugin utilise l'API d'Openai pour générer des réponses. Vous devrez configurer un compte et obtenir une clé API. L'utilisation de l'API n'est pas gratuite, mais le coût est raisonnable et dépend du nombre de jetons que vous utilisez, en termes simples, de la quantité de texte que vous envoyez et recevez (voir les prix). Notez que le plugin n'envoie aucun de votre code dans les coulisses. Vous ne partagez et ne payez que ce que vous sélectionnez spécifiquement, pour les invites et le contenu de chat.
# save api key to `~/.config/openai.token` file
echo " YOUR_OPENAI_API_KEY " > ~ /.config/openai.token
# alternatively set it as an environment variable
export OPENAI_API_KEY= " YOUR_OPENAI_API_KEY "
# or configure it with your organization id
echo " YOUR_OPENAI_API_KEY,YOUR_OPENAI_ORG_ID " > ~ /.config/openai.token
export OPENAI_API_KEY= " YOUR_OPENAI_API_KEY,YOUR_OPENAI_ORG_ID " The default api key file location is ~/.config/openai.token , but you can change it by setting the g:vim_ai_token_file_path in your .vimrc file:
let g: vim_ai_token_file_path = ' ~/.config/openai.token 'vim-plug Plug ' madox2/vim-ai ' Using built-in Vim packages :help packages
# vim
mkdir -p ~ /.vim/pack/plugins/start
git clone https://github.com/madox2/vim-ai.git ~ /.vim/pack/plugins/start/vim-ai
# neovim
mkdir -p ~ /.local/share/nvim/site/pack/plugins/start
git clone https://github.com/madox2/vim-ai.git ~ /.local/share/nvim/site/pack/plugins/start/vim-aiPour utiliser une commande AI, tapez la commande suivie d'une invite d'instructions. Vous pouvez également le combiner avec une sélection visuelle. Voici un bref aperçu des commandes disponibles:
========= Basic AI commands =========
:AI complete text
:AIEdit edit text
:AIChat continue or open new chat
============= Utilities =============
:AIRedo repeat last AI command
:AINewChat open new chat
:help vim-ai
Tip: Press Ctrl-c anytime to cancel completion
Tip: Setup your own key bindings or use command shortcuts - :AIE , :AIC , :AIR
Tip: A custom role {role} can be passed to the above commands by an initial parameter /{role}, for example :AIEdit /grammar .
Tip: Combine commands with a range :help range , for example to select the whole buffer - :%AIE fix grammar
If you are interested in more tips or would like to level up your Vim with more commands like :GitCommitMessage - suggesting a git commit message, visit the Community Wiki.
In the documentation below, <selection> denotes a visual selection or any other range, {instruction} an instruction prompt, {role} a custom role and ? symbole un paramètre facultatif.
:AI :AI - complete the text on the current line
:AI {prompt} - complete the prompt
<selection> :AI - complete the selection
<selection> :AI {instruction} - complete the selection using the instruction
<selection>? :AI /{role} {instruction}? - Utilisez le rôle pour terminer
:AIEdit <selection>? :AIEdit - edit the current line or the selection
<selection>? :AIEdit {instruction} - edit the current line or the selection using the instruction
<selection>? :AIEdit /{role} {instruction}? - Utilisez le rôle pour modifier
:AIChat :AIChat - continue or start a new conversation.
<selection>? :AIChat {instruction}? - Démarrer une nouvelle conversation compte tenu de la sélection, de l'instruction ou des deux
<selection>? :AIChat /{role} {instruction}? - Utilisez le rôle pour terminer
When the AI finishes answering, you can continue the conversation by entering insert mode, adding your prompt, and then using the command :AIChat once again.
.aichat files You can edit and save the chat conversation to an .aichat file and restore it later. Cela vous permet de créer des invites personnalisées réutilisables, par exemple:
# ./refactoring-prompt.aichat
>>> system
You are a Clean Code expert, I have the following code, please refactor it in a more clean and concise way so that my colleagues can maintain the code more easily. Also, explain why you want to refactor the code so that I can add the explanation to the Pull Request.
>>> user
[attach code]
To include files in the chat a special include role is used:
>>> user
Generate documentation for the following files
>>> include
/home/user/myproject/requirements.txt
/home/user/myproject/**/*.py
Each file's contents will be added to an additional user role message with the files separated by ==> {path} <== , where path is the path to the file. Globbing is expanded out via glob.gob and relative paths to the current working directory (as determined by getcwd() ) will be resolved to absolute paths.
Supported chat roles are >>> system , >>> user , >>> include and <<< assistant
:AINewChat :AINewChat {preset shortname}? - démarrer une nouvelle conversation
This command is used when you need to spawn a new chat in a specific way or in situation when :AIChat would normally continue conversation instead.
As a parameter you put an open chat command preset shortcut - below , tab or right . For example: :AINewChat right .
:AIRedo :AIRedo - repeat last AI command
Use this immediately after AI / AIEdit / AIChat command in order to re-try or get an alternative completion. Note that the randomness of responses heavily depends on the temperature parameter.
Dans le contexte de ce plugin, un rôle signifie une instruction et / ou une configuration d'IA réutilisables. Roles are defined in the configuration .ini file. For example by defining a grammar role:
let g: vim_ai_roles_config_file = ' /path/to/my/roles.ini ' # /path/to/my/roles.ini
[grammar]
prompt = fix spelling and grammar
[grammar.options]
temperature = 0.4 Now you can select text and run it with command :AIEdit /grammar .
Voir rôles-example.ini pour plus d'exemples.
Ce plugin ne définit aucune liaison de clé. Create your own bindings in the .vimrc to trigger AI commands, for example:
" complete text on the current line or in visual selection
nnoremap <leader> a :AI <CR>
xnoremap <leader> a :AI <CR>
" edit text with a custom prompt
xnoremap <leader> s :AIEdit fix grammar and spelling <CR>
nnoremap <leader> s :AIEdit fix grammar and spelling <CR>
" trigger chat
xnoremap <leader> c :AIChat <CR>
nnoremap <leader> c :AIChat <CR>
" redo last AI command
nnoremap <leader> r :AIRedo <CR> Chaque commande est configurée avec une variable de configuration correspondante. To customize the default configuration, initialize the config variable with a selection of options, for example put this to your .vimrc file:
let g: vim_ai_chat = {
" options " : {
" model " : " gpt-4 " ,
" temperature " : 0.2 ,
},
}Une fois l'informatique ci-dessus, vous pouvez modifier les options directement pendant la session VIM:
let g: vim_ai_chat [ ' options ' ][ ' model ' ] = ' gpt-4 '
let g: vim_ai_chat [ ' options ' ][ ' temperature ' ] = 0.2Ou personnalisez les options directement dans le tampon de chat:
[chat-options]
model =gpt-4
temperature =0.2
>>> user
generate a paragraph of lorem ipsumVous trouverez ci-dessous toutes les options de configuration disponibles, ainsi que leurs valeurs par défaut. Veuillez noter qu'il n'y a pas de limite de jeton imposée au modèle de chat.
" :AI
" - engine: complete | chat - see how to configure chat engine in the section below
" - options: openai config (see https://platform.openai.com/docs/api-reference/completions)
" - options.request_timeout: request timeout in seconds
" - options.enable_auth: enable authorization using openai key
" - options.selection_boundary: selection prompt wrapper (eliminates empty responses, see #20)
" - ui.paste_mode: use paste mode (see more info in the Notes below)
let g: vim_ai_complete = {
" engine " : " complete " ,
" options " : {
" model " : " gpt-3.5-turbo-instruct " ,
" endpoint_url " : " https://api.openai.com/v1/completions " ,
" max_tokens " : 1000 ,
" temperature " : 0.1 ,
" request_timeout " : 20 ,
" enable_auth " : 1 ,
" selection_boundary " : " ##### " ,
},
" ui " : {
" paste_mode " : 1 ,
},
}
" :AIEdit
" - engine: complete | chat - see how to configure chat engine in the section below
" - options: openai config (see https://platform.openai.com/docs/api-reference/completions)
" - options.request_timeout: request timeout in seconds
" - options.enable_auth: enable authorization using openai key
" - options.selection_boundary: selection prompt wrapper (eliminates empty responses, see #20)
" - ui.paste_mode: use paste mode (see more info in the Notes below)
let g: vim_ai_edit = {
" engine " : " complete " ,
" options " : {
" model " : " gpt-3.5-turbo-instruct " ,
" endpoint_url " : " https://api.openai.com/v1/completions " ,
" max_tokens " : 1000 ,
" temperature " : 0.1 ,
" request_timeout " : 20 ,
" enable_auth " : 1 ,
" selection_boundary " : " ##### " ,
},
" ui " : {
" paste_mode " : 1 ,
},
}
" This prompt instructs model to work with syntax highlighting
let s: initial_chat_prompt = << trim END
>>> system
You are a general assistant.
If you attach a code block add syntax type after ``` to enable syntax highlighting.
END
" :AIChat
" - options: openai config (see https://platform.openai.com/docs/api-reference/chat)
" - options.initial_prompt: prompt prepended to every chat request (list of lines or string)
" - options.request_timeout: request timeout in seconds
" - options.enable_auth: enable authorization using openai key
" - options.selection_boundary: selection prompt wrapper (eliminates empty responses, see #20)
" - ui.populate_options: put [chat-options] to the chat header
" - ui.open_chat_command: preset (preset_below, preset_tab, preset_right) or a custom command
" - ui.scratch_buffer_keep_open: re-use scratch buffer within the vim session
" - ui.paste_mode: use paste mode (see more info in the Notes below)
let g: vim_ai_chat = {
" options " : {
" model " : " gpt-4o " ,
" endpoint_url " : " https://api.openai.com/v1/chat/completions " ,
" max_tokens " : 0 ,
" temperature " : 1 ,
" request_timeout " : 20 ,
" enable_auth " : 1 ,
" selection_boundary " : " " ,
" initial_prompt " : s: initial_chat_prompt ,
},
" ui " : {
" code_syntax_enabled " : 1 ,
" populate_options " : 0 ,
" open_chat_command " : " preset_below " ,
" scratch_buffer_keep_open " : 0 ,
" paste_mode " : 1 ,
},
}
" Notes:
" ui.paste_mode
" - if disabled code indentation will work but AI doesn't always respond with a code block
" therefore it could be messed up
" - find out more in vim's help `:help paste`
" options.max_tokens
" - note that prompt + max_tokens must be less than model's token limit, see #42, #46
" - setting max tokens to 0 will exclude it from the OpenAI API request parameters, it is
" unclear/undocumented what it exactly does, but it seems to resolve issues when the model
" hits token limit, which respond with `OpenAI: HTTPError 400`Il est possible de configurer le plugin pour utiliser différents points de terminaison compatibles OpenAI. Voir certains projets sympas répertoriés dans la section API personnalisés sur le wiki communautaire.
let g: vim_ai_chat = {
" options " : {
" endpoint_url " : " http://localhost:8000/v1/chat/completions " ,
" enable_auth " : 0 ,
},
} It is possible to configure chat models, such as gpt-4o , to be used in :AI and :AIEdit commands. Ces modèles sont moins chers, mais actuellement moins adaptés à l'édition / l'achèvement du code, car ils répondent avec du texte et des commentaires de type humain.
Selon le cas d'utilisation, une bonne invite initiale peut aider à demander au modèle de chat de répondre de la manière souhaitée:
let initial_prompt = << trim END
>>> system
You are going to play a role of a completion engine with following parameter s:
Task: Provide compact code/text completion, generation, transformation or explanation
Topic: general programming and text editing
Style: Plain result without any commentary, unless commentary is necessary
Audience: Users of text editor and programmers that need to transform/generate text
END
let chat_engine_config = {
" engine " : " chat " ,
" options " : {
" model " : " gpt-4o " ,
" endpoint_url " : " https://api.openai.com/v1/chat/completions " ,
" max_tokens " : 0 ,
" temperature " : 0.1 ,
" request_timeout " : 20 ,
" selection_boundary " : " " ,
" initial_prompt " : initial_prompt,
},
}
let g: vim_ai_complete = chat_engine_config
let g: vim_ai_edit = chat_engine_configVous pourriez trouver utile une collection de commandes personnalisées sur le wiki communautaire.
To create a custom command, you can call AIRun , AIEditRun and AIChatRun functions. Par exemple:
" custom command suggesting git commit message, takes no arguments
function ! GitCommitMessageFn ()
let l: diff = system ( ' git --no-pager diff --staged ' )
let l: prompt = " generate a short commit message from the diff below: n " . l: diff
let l: config = {
" engine " : " chat " ,
" options " : {
" model " : " gpt-4o " ,
" initial_prompt " : " >>> system n you are a code assistant " ,
" temperature " : 1 ,
},
}
call vim_ai#AIRun ( l: config , l: prompt )
endfunction
command ! GitCommitMessage call GitCommitMessageFn ()
" custom command that provides a code review for selected code block
function ! CodeReviewFn ( range ) range
let l: prompt = " programming syntax is " . & filetype . " , review the code below "
let l: config = {
" options " : {
" initial_prompt " : " >>> system n you are a clean code expert " ,
},
}
exe a: firstline . " , " . a: lastline . " call vim_ai#AIChatRun(a:range, l:config, l:prompt) "
endfunction
command ! - range = 0 CodeReview <line1> , <line2> call CodeReviewFn ( <count> )Les contributions sont les bienvenues! N'hésitez pas à ouvrir une demande de traction, à signaler un problème ou à contribuer au wiki communautaire.
Précision : GPT est bon pour produire du texte et du code qui semble correct à première vue, mais qui peut être complètement faux. Assurez-vous de réviser, de lire et de tester toutes les sorties générées par ce plugin!
Confidentialité : ce plugin envoie du SMS à OpenAI lors de la génération d'achèvement et de modifications. Par conséquent, ne l'utilisez pas sur des fichiers contenant des informations sensibles.
Licence MIT
