Awesome

🚀 Ollama-chat.nvim

Chat with Ollama models directly in a Neovim buffer!

✨Features

This is a simple plugin that allows you to chat with Ollama models:

no UI to get in the way - the chat works in a normal Neovim buffer,
chat history is completely modifiable - the whole conversation is sent to the model with each new prompt,
start a chat in any open buffer,
specify chats folder and continue previous chat, start new or open a quick chat,
populate new chat buffer with selected text/code,
return to other buffers while Ollama generates a response and get a notification when the generation is complete,
stop the generation at any point using q,
chat messages are bound with User and Ollama, both surrounded with an *, making it easy to yank, delete, and change them using i*.

ollama chat

⌨️ Usage

This plugin adds the following commands that open an Ollama chat buffer:

OllamaQuickChat - opens up a quick chat in the chats_folder with the quick_chat_file name, overwriting previous chats if the file exists,
OllamaCreateNewChat - asks the user to input the chat name and creates new chat file in the chats_folder,
OllamaContinueChat - opens up Telescope to let the user choose one of previous chats in chats_folder,
alternatively open a new md buffer anywhere and use it as a chat buffer.

The chat buffer is populated with a base prompt and is completely modifiable.

If there is a selection active when the chat buffer is opened, it is copied in the new chat buffer as text, or within a corresponding code block if the source file type is code.

The Ollama model can then be prompted with the chat buffer via OllamaChat and OllamaChatCode, both of which send the entire buffer to the Ollama server, the difference being that OllamaChatCode uses the model model_code rather than model set in the opts table.

During generation you can go back to your other buffers. Once Ollama generation completes an INFO level notification will be generated to alert the user to return to the chat.

Generation can also be canceled using q.

To yank, delete, and change chat messages use i* as they are bound with *s for the User and Ollama.

The spinner animation can be disabled via the animate_spinner option. This is handy to keep the undo history clean if you want to use say undo tree.

📦 Install

First you need Ollama installed as per their instructions.

To use the plugin with lazy.nvim you can add the file lua/plugins/ollama-chat.lua:

return {
  "gerazov/ollama-chat.nvim",
  dependencies = {
    "nvim-lua/plenary.nvim",
    "stevearc/dressing.nvim",
    "nvim-telescope/telescope.nvim",
  },
}

⚙️ Configure

Here's how you can add lazy loading, some keymaps and configure options:

return {
  "gerazov/ollama-chat.nvim",
  dependencies = {
    "nvim-lua/plenary.nvim",
    "stevearc/dressing.nvim",
    "nvim-telescope/telescope.nvim",
  },
  -- lazy load on command
  cmd = {
    "OllamaQuickChat",
    "OllamaCreateNewChat",
    "OllamaContinueChat",
    "OllamaChat",
    "OllamaChatCode",
    "OllamaModel",
    "OllamaServe",
    "OllamaServeStop"
  },

  keys = {
    {
      "<leader>ocq",
      "<cmd>OllamaQuickChat<cr>",
      desc = "Ollama Quick Chat",
      mode = { "n", "x" },
      silent = true,
    },
    {
      "<leader>ocn",
      "<cmd>OllamaCreateNewChat<cr>",
      desc = "Create Ollama Chat",
      mode = { "n", "x" },
      silent = true,
    },
    {
      "<leader>occ",
      "<cmd>OllamaContinueChat<cr>",
      desc = "Continue Ollama Chat",
      mode = { "n", "x" },
      silent = true,
    },
    {
      "<leader>och",
      "<cmd>OllamaChat<cr>",
      desc = "Chat",
      mode = { "n" },
      silent = true,
    },
    {
      "<leader>ocd",
      "<cmd>OllamaChatCode<cr>",
      desc = "Chat Code",
      mode = { "n" },
      silent = true,
    },
  },

  opts = {
    chats_folder = vim.fn.stdpath("data"), -- data folder is ~/.local/share/nvim
    -- you can also choose "current" and "tmp"
    quick_chat_file = "ollama-chat.md",
    animate_spinner = true,  -- set this to false to disable spinner animation
    model = "openhermes2-mistral",
    model_code = "codellama",
    url = "http://127.0.0.1:11434",
    serve = {
      on_start = false,
      command = "ollama",
      args = { "serve" },
      stop_command = "pkill",
      stop_args = { "-SIGTERM", "ollama" },
    },
  }
}

If you want to override your Markdown @text.emphasis highlight for the User and Ollama labels, you can add the following table to your opts: