context.nvim

A Neovim plugin for AI interaction with scoped context: selected text, current line, or full file. Replies stream below your question in a floating chat window without changing your code. An optional edit mode offers code replacements with diff review.

Installation

lazy.nvim

{
  "mcncl/context.nvim",
  config = function()
    require("context").setup({
      provider = "anthropic",
      providers = {
        anthropic = {
          api_key = vim.env.ANTHROPIC_API_KEY,
          model = "claude-sonnet-4-5", -- optional
        },
      },
    })
  end,
}

packer.nvim

use {
  "mcncl/context.nvim",
  config = function()
    require("context").setup({
      provider = "anthropic",
      providers = {
        anthropic = {
          api_key = vim.env.ANTHROPIC_API_KEY,
        },
      },
    })
  end,
}

Configuration

require("context").setup({
  provider = "anthropic",  -- or "openai", "openai_compat", "amp"
  response_mode = "chat",  -- "edit" to replace code with diff review

  providers = {
    amp = {
      command = "amp",  -- Executable name or absolute path
      -- mode = "medium",  -- Optional Amp agent mode
    },
    anthropic = {
      api_key = vim.env.ANTHROPIC_API_KEY,
      model = "claude-sonnet-4-5",
      max_tokens = 4096,
    },
    openai = {
      api_key = vim.env.OPENAI_API_KEY,
      model = "gpt-5-mini",
      max_tokens = 4096,
    },
    openai_compat = {
      base_url = "http://localhost:11434/v1",  -- Ollama example
      api_key = "",  -- Optional for local
      model = "rnj-1",
      max_tokens = 4096,
    },
  },

  keymaps = {
    prompt = "<leader>ai",  -- Open prompt with context
    cancel = "<leader>ac",  -- Cancel in-flight request
  },

  ui = {
    prompt_title = "Context",
    prompt_width = 60,
    prompt_height = 3,  -- Edit-mode input height
    chat_height = 20,
  },

  diff = {
    enabled = true,    -- Edit mode only (false = instant-apply)
    vertical = true,   -- Vertical split (false = horizontal)
  },

  spinner = {
    enabled = true,
    interval = 80,       -- ms between animation frames
    done_delay = 1500,   -- ms to show done/error indicator
    minimal = false,     -- true = spinner only, no elapsed time or char count
  },
})

Amp CLI

Install the Amp CLI and run amp login in your terminal. No separate OpenAI or Anthropic API key is required. Requests use your Amp account and its usage limits or credits.

require("context").setup({ provider = "amp" })

Neovim must be able to find amp on its PATH. Otherwise, set providers.amp.command to its absolute path.

Each question starts a new local CLI process. Selected code, references, and completed chat turns go through stdin. The answer appears below the question when Amp finishes, and the process exits. Ctrl-C or closing chat terminates the active process. There is no persistent CLI session.

All tools are disabled, including file editing, shell commands, and MCP tools. The CLI runs in a temporary directory with isolated settings and IDE context disabled. Your personal and repository Amp settings are not loaded or changed; use providers.amp.mode to select an agent mode. Your saved Amp login is reused. Amp stores each request as a private thread and archives it after execution; closing the Neovim chat does not delete those threads.

Usage

Keymaps

Mode	Key	Action
n, v	<leader>ai	Open prompt with context
n	<leader>ac	Cancel active request

Commands

• :Context - Open prompt with auto-detected context
• :ContextVisual - Open prompt with visual selection
• :ContextLine - Open prompt with current line
• :ContextFile - Open prompt with full file
• :ContextCancel - Cancel active request
• :ContextDiffAccept - Accept the proposed diff
• :ContextDiffReject - Reject the proposed diff
• :ContextProvider [name] - Get or set current provider

Chat

Open a prompt with <leader>ai or a :Context command, type your question, and press Enter. The answer streams below the question. Your source buffer is not changed, even when the reply contains code.

Each follow-up includes the previous completed questions and answers. The selected code is captured when you open the chat; @path/to/file.lua references add files to individual questions. Only one request runs at a time. You can type your next question while waiting, then submit it after the reply finishes.

Key	Action
Enter (insert mode)	Send a question, or accept visible @file completion
Escape	Leave insert mode to scroll or copy replies
Enter (normal mode)	Return to the input line
Ctrl-C (insert mode)	Cancel the current reply without closing chat
q (normal mode)	Close chat and cancel any active reply

Closing the window discards the conversation. Opening another prompt while chat is open focuses the existing conversation. To use different selected code, close the chat and open a new one.

Use chat_system_prompt to customize chat instructions. The existing system_prompt setting applies to edit mode.

Diff Preview

Set response_mode = "edit" to use the code-replacement workflow. When a response completes, a side-by-side diff opens showing the original code vs the proposed replacement. From either window:

Key	Action
<CR>	Accept the change
q	Reject and restore original

The buffer is locked during review to prevent edits. Closing the diff window also rejects. To disable the preview and apply changes instantly, set diff = { enabled = false }.

Context Detection

1. Visual mode: Select text, then <leader>ai - uses selection
2. Normal mode at line 1: <leader>ai - uses entire file
3. Normal mode elsewhere: <leader>ai - uses current line

Statusline Integration

Expose streaming/review status in your statusline (e.g. lualine):

lualine_x = {
  { require("context").get_status_line, cond = require("context").is_active },
}

Programmatic API

local context = require("context")

context.prompt()           -- Open prompt with auto-detected context
context.prompt_visual()    -- Force visual mode context
context.prompt_line()      -- Force line mode context
context.prompt_file()      -- Force file mode context
context.cancel()           -- Cancel in-flight request
context.set_provider(name) -- Switch provider at runtime
context.is_active()        -- Check if request is in progress
context.diff_accept()      -- Accept active diff review
context.diff_reject()      -- Reject active diff review
context.get_status()       -- Structured status for programmatic use
context.get_status_line()  -- Formatted string for statusline

Testing

Run the chat and edit-mode regression checks without API credentials:

nvim --headless -u NONE -l tests/chat.lua
nvim --headless -u NONE -l tests/amp.lua
luacheck lua/ tests/ --globals vim

License

MIT