codegpt-ng.nvim

codegpt-ng is a minimalist plugin for neovim that provides commands to interact with AI backend. The focus is around code related usages. So code completion, refactorings, generating docs, etc.

This is a fork of the original CodeGPT repository from github user @dpayne. All credit goes to him for the initial work.

This fork does the following:

• Full support for Ollama API
• New table-based configuration instead of global variables
• Streaming mode for real-time popup responses
• Refactored for idiomatic Lua and neovim plugin style
• Simplified command system with explicit configuration
• Tests with plenary library

Although this fork introduces breaking changes and a complete rewrite, I've tried to preserve the original project's minimalist ethos — a tool that connects to AI APIs without getting in the way. The goal remains to provide simple, code-focused interactions that stay lightweight and unobtrusive, letting developers leverage AI power while maintaining control over their workflow.

Installation

• The plugins 'plenary' and 'nui' are also required.

Installing with Lazy.

{
  "blob42/codegpt-ng.nvim",
  dependencies = {
    'nvim-lua/plenary.nvim',
    'MunifTanjim/nui.nvim',
  },
  opts = {
    -- API configuration
    api = {
      provider = "openai",  -- or "Ollama", "Azure", etc.
      openai_api_key = os.getenv("OPENAI_API_KEY"),
      chat_completions_url = "https://api.openai.com/v1/chat/completions",
    },
    
    -- Command defaults
    commands = {
      completion = {
        model = "gpt-3.5-turbo",
        user_message_template = "This is a template...",
        callback_type = "replace_lines",
      },
      tests = {
        language_instructions = { java = "Use TestNG framework" },
      },
      -- Add custom commands here
    },
    
    -- UI configuration
    ui = {
      popup_type = "popup",  -- or "horizontal", "vertical"
      text_popup_filetype = "markdown",
      commands = {
        quit = "q",
        use_as_output = "<c-o>",
      },
      popup_options = {
        relative = "editor",
        position = "50%",
        size = { width = "80%", height = "80%" },
      },
      popup_border = { style = "rounded" },
      popup_window_options = { wrap = true, number = true },
    },
    
    -- Status hooks
    hooks = {
      request_started = function() vim.cmd("hi StatusLine ctermfg=yellow") end,
      request_finished = function() vim.cmd("hi StatusLine ctermfg=NONE") end,
    },
    
    -- Other options
    clear_visual_selection = true,
  }
}

Installing with Packer.

use({
   "blob42/codegpt-ng.nvim",
   requires = {
      "MunifTanjim/nui.nvim",
      "nvim-lua/plenary.nvim",
   },
   config = function()
      require("codegpt").setup({
        -- Configuration here
      })
   end
})

Commands

The top-level command is :Chat. The behavior is different depending on whether text is selected and/or arguments are passed.

You can also use :VChat to temporary enforce the vertical layout.

Completion

• :Chat with text selection will trigger the completion command, ChatGPT will try to complete the selected code snippet.

Code Edit

• :Chat some instructions with text selection and command args will invoke the code_edit command.

Code Edit

• :Chat <command> if there is only one argument and that argument matches a command, it will invoke that command with the given text selection. In the below example :Chat tests will attempt to write units for the selected code.

Chat

• :Chat hello world without any text selection will trigger the chat command.

A full list of predefined commands are below

command	input	Description
completion	text selection	Will ask ChatGPT to complete the selected code.
code_edit	text selection and command args	Will ask ChatGPT to apply the given instructions (the command args) to the selected code.
explain	text selection	Will ask ChatGPT to explain the selected code.
question	text selection	Will pass the commands args to ChatGPT and return the answer in a text popup.
debug	text selection	Will pass the code selection to ChatGPT analyze it for bugs, the results will be in a text popup.
doc	text selection	Will ask ChatGPT to document the selected code.
opt	text selection	Will ask ChatGPT to optimize the selected code.
tests	text selection	Will ask ChatGPT to write unit tests for the selected code.
chat	command args	Will pass the given command args to ChatGPT and return the response in a popup.

Configuration

Global Configuration

require("codegpt").setup({
  api = {
    provider = "openai",  -- or "Ollama", "Azure", etc.
    openai_api_key = os.getenv("OPENAI_API_KEY"),
    chat_completions_url = "https://api.openai.com/v1/chat/completions",
  },
  commands = {
    -- Command defaults
  },
  ui = {
    -- UI configuration
  },
  hooks = {
    -- Status hooks
  },
  clear_visual_selection = true,
})

Overriding Command Configurations

The configuration table commands can be used to override command configurations.

commands = {
  completion = {
    model = "gpt-3.5-turbo",
    user_message_template = "This is a template...",
    callback_type = "replace_lines",
  },
  tests = {
    language_instructions = { java = "Use TestNG framework" },
  }
}

Custom Commands

Custom commands can be added to the commands table.

commands = {
  modernize = {
    user_message_template = "Modernize the code...",
    language_instructions = { cpp = "..." }
  }
}

UI Configuration

ui = {
  popup_type = "popup",  -- or "horizontal", "vertical"
  text_popup_filetype = "markdown",
  commands = {
    quit = "q",
    use_as_output = "<c-o>",
  },
  popup_options = {
    relative = "editor",
    position = "50%",
    size = { width = "80%", height = "80%" },
  },
  popup_border = { style = "rounded" },
  popup_window_options = { wrap = true, number = true },
}

Status Hooks

hooks = {
  request_started = function() vim.cmd("hi StatusLine ctermfg=yellow") end,
  request_finished = function() vim.cmd("hi StatusLine ctermfg=NONE") end,
}

Templates

The system_message_template and user_message_template can contain template macros. For example:

macro	description
{{filetype}}	The filetype of the current buffer.
{{text_selection}}	The selected text in the current buffer.
{{language}}	The name of the programming language in the current buffer.
{{command_args}}	Everything passed to the command as an argument, joined with spaces.
{{language_instructions}}	The found value in the language_instructions map.

Callback Types

name	Description
replace_lines	replaces the current lines with the response. If no text is selected it will insert the response at the cursor.
text_popup	Will display the result in a text popup window.
code_popup	Will display the results in a popup window with the filetype set to the filetype of the current buffer

Example Configuration

require("codegpt").setup({
  api = {
    provider = "openai",
    openai_api_key = os.getenv("OPENAI_API_KEY"),
  },
  models = {
    ollama = {
      default = 'qwen3:4b',
      ['qwen3:4b'] = {
        max_tokens = 8192,
        temperature = 0.8,
        append_string = '/no_think',
      },
  },
  commands = {
    tests = {
      language_instructions = { java = "Use TestNG framework" },
    },
    doc = {
      language_instructions = { python = "Use Google style docstrings" },
      max_tokens = 1024,
    },
    code_edit = {
      system_message_template = "You are {{language}} developer.",
      user_message_template = "I have the following {{language}} code: ```{{filetype}}\n{{text_selection}}```\nEdit the above code. {{language_instructions}}",
      callback_type = "code_popup",
    },
    modernize = {
      user_message_template = "I have the following {{language}} code: ```{{filetype}}\n{{text_selection}}```\nModernize the above code. Use current best practices. Only return the code snippet and comments. {{language_instructions}}",
      language_instructions = {
        cpp = "Use modern C++ syntax. Use auto where possible. Do not import std. Use trailing return type. Use the c++11, c++14, c++17, and c++20 standards where applicable.",
      },
    }
  },
  ui = {
    popup_type = "popup",
    text_popup_filetype = "markdown",
    commands = {
      quit = "q",
      use_as_output = "<c-o>",
    },
    popup_options = {
      relative = "editor",
      position = "50%",
      size = { width = "80%", height = "80%" },
    },
    popup_border = { style = "rounded" },
    popup_window_options = { wrap = true, number = true },
  },
  hooks = {
    request_started = function() vim.cmd("hi StatusLine ctermfg=yellow") end,
    request_finished = function() vim.cmd("hi StatusLine ctermfg=NONE") end,
  },
  clear_visual_selection = true,
})