🤖 Enlighten

Turn Neovim into an AI Code Editor

Edit	Chat
Enlighten.Edit.mov	Enlighten.Chat.mov

🔦 Philosophy

There are lots of agentic solutions out there. This is not one of them. You still often need to make a change in the editor. This plugin will meet you where you are and then get out of the way.

🥖 Features

• Generate code from a prompt.
• Edit selected code in place with change highlights.
• Conversational chat without leaving Neovim.
• Conversation and prompt history.
• Edit buffer from chat with the conversation as context.
• Use other tools for reviewing, accepting and rejecting generated code by turning highlights off.
• All OpenAI and Anthropic models can be supported (I can't always keep up though).
• curl is the only dependency!

💾 Setup

Must have Neovim 0.10.0+

curl is required.

You will also need to set the environment variable OPENAI_API_KEY or ANTHROPIC_API_KEY depending on the AI provider you are using.

Installation with lazy.nvim

  {
    "evanmcneely/enlighten.nvim",
    event = "VeryLazy",
    opts = {
      -- ...
    },
    keys = {
      { "<leader>ae", function() require("enlighten").edit() end, desc = "Edit", mode = { "n", "v" } },
      { "<leader>ac", function() require("enlighten").chat() end, desc = "Chat", mode = { "n", "v" } },
      { "<leader>ay", function() require("enlighten").keep() end, desc = "Keep change", mode = { "n", "v" } },
      { "<leader>aY", function() require("enlighten").keep_all() end, desc = "Keep all changes", mode = "n" },
      { "<leader>an", function() require("enlighten").discard() end, desc = "Discard change", mode = { "n", "v" } },
      { "<leader>aN", function() require("enlighten").discard_all() end, desc = "Discard all changes", mode = "n" },
    },
  },

⚙️ Configuration

This is the default configuration.

  {
    ai = {
      provider = "openai", -- AI provider. Only "openai" or "anthropic" or supported.
      model = "gpt-4o", -- model name for the specified provider. Only chat completion models are supported (plus the o3-mini reasoning model)
      temperature = 0,
      tokens = 4096,
      timeout = 120, -- recommended to keep very high
    },
    settings = {
      context = 500, -- lines above and below the selected text passed to the model as context
        -- Can be "diff" or "change" or "off":
        -- - "diff" will show added and removed lines with DiffAdd and DiffRemove highlights
        -- - "change" when a hunk has both added and removed lines, this will show only generated content with DiffText highlights
        -- - "off" will not add highlights around generated content (use other tooling to review, accept and revert changes)
      diff_mode = "diff",
      edit = {
        width = 80, -- prompt window width
        height = 5, -- prompt window height
        showTitle = true, -- show the title in the prompt window
        showHelp = true, -- show the help footer in the prompt window
        border = "═", -- top/bottom border character of prompt window
        auto_close = false, -- close the prompt on submit and auto-accept generated content
      },
      chat = {
        width = 80, -- chat window width
        split = "right", -- split the chat window 'left' or 'right'
      },
    },
  }

To customize this configuration, pass overrides into setup: require('enlighten').setup({...}) .

Using Anthropic as the provider

Example configuration to use Anthropic AI as the completion provider:

  {
    ai = {
      provider = "anthropic",
      model = "claude-sonnet-4-6",
    },
  }

Feature specific configuration

You can override the AI and Settings configurations for the edit or chat feature using the following feature specific overrides:

  {
    ai = {
      edit = {
        timeout = 30, -- set a lower timeout for the prompt feature only
      },
      chat = {
        model = "o3-mini", -- use a different model for the chat feature only
      }
    },
    settings = {
      edit = {
        -- use minimalist UI
        height = 3,
        showTitle = false,
        showHelp = false,
        border = "",
      },
      chat = {
        diff_mode = "off" -- turn diff highlights off when editing from chat
      }
    }
  }

These are just examples, base settings and all ai configurations can be overridden in this way.

📖 Usage

Prompt completion

Select the code you want to edit (from normal mode, the line under the cursor is considered selected). Open the prompt and write your instructions. The keymaps available to you are:

• <CR> - submit the prompt for completion
• q - close the prompt window (clears any generated code)
• <C-y> - accept the generated code
• <C-o> - scroll back through past prompts
• <C-i> - scroll forward through past prompts

Generated code is diff'd against the initially selected code and highlighted as green (add) or red (remove) appropriately.

Runtime options

enlighten.edit() accepts an optional table to override edit settings at call time. This lets you set up different keybindings for different workflows:

-- Standard edit with diff review
vim.keymap.set({ "n", "v" }, "<leader>ae", function() require("enlighten").edit() end)

-- Fire and forget: prompt closes on submit, content streams in, changes auto-accepted
vim.keymap.set({ "n", "v" }, "<leader>aE", function()
  require("enlighten").edit({ diff_mode = "off", auto_close = true })
end)

When auto_close is true, the prompt window closes immediately on submit. If diff_mode is "off", changes are also auto-accepted. If diff_mode is "diff" or "change", the prompt still closes but diff highlights remain in the buffer so you can review and accept or discard individual changes.

Any edit setting (diff_mode, auto_close, width, height, etc.) can be overridden this way.

Chat

Open the chat window. You can optionally select code from the buffer to have it populate the prompt. The keymaps available to you are:

• <CR> - submit the prompt for completion (normal mode)
• <C-CR> - submit the prompt for completion (insert mode)
• <S-C-CR - AI edit buffer with context from chat (⚠️ Experimental)
• q - close the chat window
• <C-o> - scroll back through past chat conversations
• <C-i> - scroll forward through past chat conversations
• <C-x> - stop the streamed response

Chat responses are streamed into the chat buffer.

@Mentions

⚠️ Experimental

Dependant on nvim-cmp and plenary.nvim.

While using either of the above features, type @ to see a list of available completions.

• @files - Uses vim.ui.select to open a project file picker. Selected file contents are added to the chat/prompt.
• @target - Adds the target buffer content (the buffer your cursor was in when the chat/prompt was initiated) to the chat/prompt.
• @buffers - Adds all open buffers content to the chat/prompt.
• @quickfix - Adds all buffer content from the quickfix list to the chat/prompt.
• @enlighten - Uses vim.ui.select to open a picker with files from a .enlighten/ directory in the project root and home directories. Selected file contents are added to the chat/prompt. This enables reusable custom prompts, instructions or documentation.

If you have not configured vim.ui.select, try snacks.nvim and this simple configuration to make life easy.

  {
    "folke/snacks.nvim",
    lazy=false,
    opts = {
      picker = {
        ui_select = true, -- only use for vim.ui.select
      },
    },
  },

👍 Kudos

• aduros/ai.vim : Which provided the foundation of this project. The two projects look nothing the same now, but ai.vim made me think this was possible.

• Cursor : The AI Code Editor which, in my opinion, has pioneered AI-editor integrations and inspired many features here.

• avante.nvim : A very different experience and ultimate goal, but with a lot of overlap in problems and implementation.

• spinner.nvim : used in this project.