coddit – with LLMs

coddit.nvim is a Neovim plugin to seamlessly harness coding capabilities of LLMs on your projects. Just select the code and ask away.

(α) You can now also ask ChatGPT to assist you while coding, directly in Neovim, via coddit.py and coddit Chrome extension.

Showcase

demo1.mov

demo2.mov

Setup

The following guide is specific to the lazy.nvim plugin manager.

Installation

Add the following table to your lazy.nvim configuration for a default setup using Anthropic's Claude Sonnet 3.5.

{
  "presindent/coddit.nvim",
  lazy = false,
  branch = "main",
}

Customization

For custom model configurations, use an expanded setup such as this:

{
  "presindent/coddit.nvim",
  lazy = false,
  config = function()
    require("coddit").setup({
      models = {
        ["haiku"] = {
          model = "claude-3-haiku-20240307",
          api_type = "anthropic",
        },
        ["sonnet"] = {
          model = "claude-3-5-sonnet-latest",
          api_type = "anthropic",
        },
        ["gpt-4o"] = {
          model = "gpt-4o",
          api_type = "openai",
        },
      },
      selected_model = "sonnet",  -- set this to "gpt-4o" to use OpenAI's GPT-4O
      max_tokens = 1024,
      anthropic_version = "2023-06-01",
      stream = true,  -- set this to false to disable streaming
      show_diff = true,  -- set this to false to skip diffview
    })
  end,
}

Note: coddit.nvim uses OPENAI_API_KEY or ANTHROPIC_API_KEY environment variables by default for API keys. You may custom keys using api_key attribute in a model table during the setup.

You can also define custom API types with their own headers and API payloads by defining get_headers and get_api_payload functions. You will also have to define an extract_assistant_response function to extract the assistants response from the API's JSON response. You can also define these functions for individual models. Here's an example defining ollama API type and adding deepseek-coder:6.7b to your options.

local util = require("coddit.util")

require("coddit").setup({
  api_types = {
    ["ollama"] = {
      endpoint = "http://localhost:11434/api/chat",
      get_headers = function(_)
        return { ["Content-Type"] = "application/json" }
      end,
      get_api_payload = function(prompt, model_opts, api_opts, module_opts)
        local system_prompt = model_opts.system_prompt or api_opts.system_prompt or module_opts.system_prompt
        local max_tokens = model_opts.max_tokens or api_opts.max_tokens or module_opts.max_tokens
        local stream = util.get_first_boolean(true, { model_opts.stream, api_opts.stream, module_opts.stream })

        return vim.fn.json_encode({
          model = model_opts.model,
          stream = stream,
          options = {
            num_predict = max_tokens,
          },
          messages = {
            { role = "system", content = system_prompt },
            { role = "user", content = prompt },
          },
        })
      end,
      extract_assistant_response = function(response)
        local decoded = vim.fn.json_decode(response)
        if decoded and decoded.message and decoded.message.content then
          return decoded.message.content
        end
        return nil
      end,
      extract_text_delta = function(response)
        local decoded = vim.fn.json_decode(response)
        if decoded and decoded.message and decoded.message.content then
          return decoded.message.content
        end
        return nil
      end,
      stream = true,
    },
  },
  models = {
    ["deepseek-coder"] = {
      model = "deepseek-coder:6.7b",
      api_type = "ollama",
    },
  },
  selected_model = "sonnet",
})

Usage

Lua API

local coddit = require("coddit")

-- Configure the module.
---@param opts? Opts Table of configuration options (optional)
coddit.setup({ ... })

-- Invoke the selected LLM endpoint.
-- If a visual block is selected, the snippet is communicated to the LLM,
-- else, the part of the buffer above and including the current cursor row
-- is communicated.
---@param toggle_show_diff? boolean Skip the diff if false, not if unspecified
coddit.call(toggle_show_diff)

-- Select the model to call when `coddit.call` is invoked. It updates the
-- `coddit.selected_model` attribute, which may also be specified during
-- the initial setup.
--
-- If invoked without the `model_name` argument, it displays a menu to pick
-- the model from.
---@param model_name? string
coddit.select_model(model_name)

-- Close the diff view if active.
coddit.close_diff_view()

Recommended Mappings

These keymaps provide convenient access to coddit's main functions. Note that these keymaps are not set by default, you will have to add the following snippet to the config function in lazy.nvim config, or in another appropriate configuration file:

vim.keymap.set({ "n", "v" }, "<leader>aic", require("coddit").call, { desc = "Invoke coddit" })
vim.keymap.set({ "n", "v" }, "<leader>aid", require("coddit").close_diff_view, { desc = "Close coddit diffview" })

vim.keymap.set("n", "<leader>aim", function()
  require("coddit").select_model()
end, { desc = "Select model for coddit" })

vim.keymap.set("v", "<leader>aim", function()
  require("coddit").select_model(nil, function()
    vim.cmd("normal! gv")
    require("coddit").call()
  end)
end, { desc = "Select model for coddit and call" })

vim.keymap.set({ "n", "v" }, "<leader>aiC", function()
  require("coddit").call(true)
end, { desc = "Invoke coddit with toggled diff view" })

Selecting the Model

The API and the keymap discussed above facilitate selecting your favourite LLM on the go.

Roadmap

• Implement actions and CLI interaction.
• Tracking other enhancements and fixes as issues.

Similar Plugins

This plugin is inspired by llm.nvim by melbaldove.