- Notifications
You must be signed in to change notification settings - Fork116
Neovim plugin to generate text using LLMs with customizable prompts
License
David-Kunz/gen.nvim
Folders and files
| Name | Name | Last commit message | Last commit date | |
|---|---|---|---|---|
Repository files navigation
Generate text using LLMs with customizable prompts
Install with your favorite plugin manager, e.g.lazy.nvim
Example with Lazy
-- Minimal configuration{"David-Kunz/gen.nvim"},
-- Custom Parameters (with defaults){"David-Kunz/gen.nvim",opts= {model="mistral",-- The default model to use.quit_map="q",-- set keymap to close the response windowretry_map="<c-r>",-- set keymap to re-send the current promptaccept_map="<c-cr>",-- set keymap to replace the previous selection with the last resulthost="localhost",-- The host running the Ollama service.port="11434",-- The port on which the Ollama service is listening.display_mode="float",-- The display mode. Can be "float" or "split" or "horizontal-split".show_prompt=false,-- Shows the prompt submitted to Ollama. Can be true (3 lines) or "full".show_model=false,-- Displays which model you are using at the beginning of your chat session.no_auto_close=false,-- Never closes the window automatically.file=false,-- Write the payload to a temporary file to keep the command short.hidden=false,-- Hide the generation window (if true, will implicitly set `prompt.replace = true`), requires Neovim >= 0.10init=function(options)pcall(io.popen,"ollama serve > /dev/null 2>&1 &")end,-- Function to initialize Ollamacommand=function(options)localbody= {model=options.model,stream=true}return"curl --silent --no-buffer -X POST http://"..options.host..":"..options.port.."/api/chat -d $body"end,-- The command for the Ollama service. You can use placeholders $prompt, $model and $body (shellescaped).-- This can also be a command string.-- The executed command must return a JSON object with { response, context }-- (context property is optional).-- list_models = '<omitted lua function>', -- Retrieves a list of model namesresult_filetype="markdown",-- Configure filetype of the result bufferdebug=false-- Prints errors and the command which is run. }},
Here are allavailable models.
Alternatively, you can call thesetup function:
require('gen').setup({-- same as above})
Use commandGen to generate text based on predefined and customizable prompts.
Example key maps:
vim.keymap.set({'n','v'},'<leader>]',':Gen<CR>')
You can also directly invoke it with one of thepredefined prompts or your custom prompts:
vim.keymap.set('v','<leader>]',':Gen Enhance_Grammar_Spelling<CR>')
After a conversation begins, the entire context is sent to the LLM. That allows you to ask follow-up questions with
:GenChat
and once the window is closed, you start with a fresh conversation.
For prompts which don't automatically replace the previously selected text (replace = false), you can replace the selected text with the generated output with<c-cr>.
You can select a model from a list of all installed models with
require('gen').select_model()
All prompts are defined inrequire('gen').prompts, you can enhance or modify them.
Example:
require('gen').prompts['Elaborate_Text']= {prompt="Elaborate the following text:\n$text",replace=true}require('gen').prompts['Fix_Code']= {prompt="Fix the following code. Only output the result in format ```$filetype\n...\n```:\n```$filetype\n$text\n```",replace=true,extract="```$filetype\n(.-)```"}
You can use the following properties per prompt:
prompt: (string | function) Prompt either as a string or a function which should return a string. The result can use the following placeholders:$text: Visually selected text or the content of the current buffer$filetype: File type of the buffer (e.g.javascript)$input: Additional user input$register: Value of the unnamed register (yanked text)
replace:trueif the selected text shall be replaced with the generated outputextract: Regular expression used to extract the generated resultmodel: The model to use, default:mistral
User selections can be delegated toTelescope withtelescope-ui-select.