LLM Plugin for Neovim

A Neovim plugin that integrates the llm command-line tool for AI-powered text generation directly into your editor.

Features

• Asynchronous execution - Run LLM commands without blocking your editor
• Synchronous mode - Insert LLM output directly at cursor position with ! bang
• Visual selection support - Send selected text as input to the LLM
• Dedicated output window - View LLM responses in a persistent split window
• Progress indicator - Animated feedback while waiting for responses
• Vim modifiers support - Use %, %:p, and other filename modifiers in commands
• Smart output formatting - Automatically formats output with comment strings when inserting into code files

Prerequisites

You must have the llm CLI tool installed and configured:

uv tool install llm
# or pipx install llm

Refer to their docs for more information

Installation

Using lazy.nvim:

{
    {
        "https://github.com/BlakeJC94/llm.nvim",
        opts = {
            autoscroll = true,
            llm_path = "llm",  -- Can be a string or a function
            template = nil,     -- Default template name, passed as -t <template>
            split = {
                direction = "horizontal",
                size = 14,
                position = "bottom",
            },
        },
        cmd = {
            "LLM",
            "LLMToggle",
            "LLMOpen",
            "LLMClose",
            "LLMStop",
        },
        -- Optional: set keymaps
        keys = {
            {
                "<Leader>s",
                ":LLM ",
                mode = {"n", "v"},
            },
            {
                "<Leader>S",
                ":LLMToggle<CR>",
                mode = "n",
            },

    },
}

Commands

:LLM {args}

Execute an LLM command asynchronously and display output in the LLM window.

Arguments that don't start with a known llm subcommand (like logs, templates, chat, etc.) are automatically prefixed with prompt, so :LLM "Explain this" runs llm prompt "Explain this". If a template is configured, -t <template> is also prepended.

:LLM "Explain this code" < %
:LLM "Write a function to reverse a string"
:'<,'>LLM "Refactor this code"
:LLM logs -n 1

:LLM! {args}

Execute an LLM command synchronously and insert output at the cursor position. Output is automatically wrapped in comments based on the current filetype.

:LLM! "Add error handling"
:'<,'>LLM! "Add JSDoc comments"

:LLMToggle

Toggle the LLM output window open/closed.

:LLMOpen

Open the LLM output window.

:LLMClose

Close the LLM output window.

:LLMStop

Stop the currently running LLM command.

Configuration

The plugin can be configured during setup:

opts = {
    template = "default",          -- default template passed as -t <template>
    split = {
        direction = "horizontal",   -- or "vertical"
        size = 16,                  -- window size in lines/columns
        position = "bottom",        -- "bottom", "top", "left", or "right"
    },
    wo = {
        -- Window options for the LLM window
        wrap = true,
        number = false,
        relativenumber = false,
        cursorline = false,
        cursorcolumn = false,
        signcolumn = "no",
        spell = false,
    },
    bo = {
        -- Buffer options for the LLM buffer
        buflisted = false,
        filetype = "markdown",
    },
})

Templates

The template option sets a default template passed to llm via the -t flag on every :LLM invocation. When nil or empty, the plugin will fall back to the LLM_DEFAULT_TEMPLATE environment variable if set. If neither is provided, no -t flag is added.

Templates are YAML files stored in the directory returned by llm templates path. For example:

# $(llm templates path)/default.yaml

name: default
model: openrouter/deepseek/deepseek-v4-flash

system: |
  Be as short, direct and concise as possible. Do not make any further
  suggestions --- I'll ask exactly what I need and nothing more. If I need some
  coding help, just return the required code with minimal explanation. Again --- I
  will ask you if I need you to explain it further. Lists are preferred, and
  please no bold text

With template = "default" in your config, every :LLM call will automatically use -t default, selecting the model and system prompt defined above.

Usage Examples

Ask a question about current file

:LLM "Explain what this file does" < %

Continue from last conversation

:LLM -c "Explain this part in more detail"

Refactor selected code

Select code in visual mode, then:

:'<,'>LLM "Refactor this to be more efficient"

Generate code at cursor

:LLM! "Write a function that validates email addresses"

Print the last response (requires jq)

:LLM logs -n 1 --json | jq -r '.[].response'

Add documentation to selection

Select a function, then:

:'<,'>LLM! "Add comprehensive docstrings and comments"

The output will be automatically formatted as comments based on your filetype.

License

MIT