A Neovim plugin for counting tokens in text files using various AI model tokenizers. Features background caching for optimal performance with file explorers and status lines.

{
  "3ZsForInsomnia/token-count.nvim",
  opts = {
    model = "gpt-5",      -- Default model for counting
  },
  config = true,
}

The plugin automatically creates a virtual environment and installs required Python libraries when first used:

:TokenCount  " Triggers automatic setup on first use

Check setup status:

:checkhealth token-count

:TokenCount       " Count tokens in current buffer
:TokenCountModel  " Change the active model
:TokenCountAll    " Count tokens across all open buffers
:TokenCountSelection " Count tokens in visual selection (in visual mode)

require("token-count").setup({
  model = "gpt-4",                  -- Default model (see MODELS.md for all options)
  log_level = "warn",               -- Logging verbosity
  context_warning_threshold = 0.4,  -- Warn at 40% context usage
  
  -- Ignore patterns for background processing (Lua patterns)
  ignore_patterns = {
    "node_modules/.*",   -- Node.js dependencies
    "%.git/.*",          -- Git repository files
    "%.svn/.*",          -- SVN files
    "build/.*",          -- Build directories
    "dist/.*",           -- Distribution directories
    "target/.*",         -- Build target directories
    "vendor/.*",         -- Vendor dependencies
    "%.DS_Store",        -- macOS system files
    "%.vscode/.*",       -- VS Code settings
    "%.idea/.*",         -- IntelliJ settings
  },
  
  -- Optional: Enable official API token counting (requires API keys)
  enable_official_anthropic_counter = false, -- Requires ANTHROPIC_API_KEY
  enable_official_gemini_counter = false,    -- Requires GOOGLE_API_KEY
})

The plugin automatically detects and processes a wide variety of text-based files including:

• Programming languages: JavaScript, TypeScript, Python, Go, Rust, Java, C/C++, Lua, Ruby, PHP, Swift, Kotlin, and many more
• Web technologies: HTML, CSS, SCSS, Vue, Svelte, JSON, XML
• Documentation: Markdown, reStructuredText, LaTeX, plain text
• Configuration: YAML, TOML, INI, environment files
• Data formats: CSV, SQL, GraphQL
• Infrastructure: Dockerfile, Terraform, Kubernetes YAML

• Files under 512KB: Full token counting
• Files over 512KB: Displayed as "LARGE" to indicate the file exceeds processing limits
• Performance: Background processing uses parallel execution for optimal speed

The plugin automatically skips common directories and files that shouldn't be processed:

• node_modules, .git, build, dist, target directories
• System files like .DS_Store
• IDE configuration directories

You can customize ignore patterns in your configuration (see Configuration section above).

📋 View Complete Models List →

The plugin supports 60+ models including GPT-4/5, Claude, Gemini, Llama, Grok, and more. Token counting accuracy varies by provider:

If Telescope is installed, you get an enhanced model picker with fuzzy search and preview:

:TokenCountModel  " Opens Telescope picker automatically

Or use directly:

:Telescope token_count models

The picker shows:

require('lualine').setup({
  sections = {
    lualine_c = { 
      require('token-count.integrations.lualine').current_buffer 
    }
  },
  winbar = {
    lualine_c = { 
      require('token-count.integrations.lualine').all_buffers 
    }
  }
})

require("token-count.integrations.neo-tree").setup({
  component = {
    enabled = true,
    show_icon = true,
    icon = "🪙",
  }
})

Shows token counts next to files and directories with background processing.

Select text in visual mode and use :TokenCountSelection:

vim.keymap.set("v", "<leader>tc", ":TokenCountSelection<CR>", {
    desc = "Count tokens in visual selection",
    silent = true
})

require("token-count").get_current_buffer_count(function(result, error)
  if result then
    print("Tokens:", result.token_count)
    print("Model:", result.model_config.name)
  end
end)

local models = require("token-count").get_available_models()

local model_config = require("token-count").get_current_model()

local cache = require("token-count.cache")

local file_tokens = cache.get_file_token_count("/path/to/file.lua")

cache.clear_cache()
local stats = cache.get_stats()

:checkhealth token-count

Provides comprehensive status of:

The plugin features a unified background cache that:

Cache is enabled by default with sensible settings. See ADVANCED.md for detailed configuration.

:TokenCountVenvStatus  " Check detailed status
:TokenCountVenvSetup   " Recreate if needed

Ensure Python 3.7+ is available:

python3 --version

Check available models:

:TokenCountModel  " Browse and select models

MIT License - see LICENSE file for details.

Count AI model tokens in your files. Works locally for most models, with smart background caching that stays out of your way.

• Know if your code fits in model context windows before you hit limits
• Background counting doesn't slow down your editor - processes files when you're not typing
• Exact counts for OpenAI and DeepSeek models, smart estimates for everything else
• Seamless integrations with lualine and neo-tree show counts without extra commands
• Large file handling - estimates large background files (marked with *), full counts for active files

{
  "zacharylevinw/token-count.nvim",
  dependencies = {
    "nvim-telescope/telescope.nvim", -- Optional: enhanced model selection
  },
  config = function()
    require("token-count").setup({
      model = "gpt-4o", -- Default model
    })
  end,
}

Prerequisites: Neovim 0.9.0+, Python 3.7+

The plugin automatically sets up its Python environment and dependencies on first use. Just run :TokenCount and it handles the rest.

:TokenCount        " Count tokens in current file
:TokenCountModel   " Switch between models
:TokenCountAll     " Count all open files

require('lualine').setup({
  sections = {
    lualine_c = { 
      require('token-count.integrations.lualine').current_buffer 
    }
  }
})

require("token-count.integrations.neo-tree").setup({
  component = {
    enabled = true,
    show_icon = true,
    icon = "🪙",
  }
})

Shows token counts next to files and directories. Large files in the background get estimated counts (marked with *).

• Active/small files: Full accurate counting using the best available method
• Large background files (>512KB): Smart estimation to keep things fast
• Exact counting: OpenAI models (via tiktoken), DeepSeek models (via official tokenizer)
• Smart estimates: All other models via tokencost library
• Optional API counting: Set ANTHROPIC_API_KEY or GOOGLE_API_KEY for exact Anthropic/Google counts (not recommended - prefer local)

Supports 60+ models including GPT-4/5, Claude, Gemini, Llama, and more. See MODELS.md for the complete list.

Switch models anytime with :TokenCountModel (uses Telescope if available for better search).

See ADVANCED.md for:

• Public API for custom integrations
• Using with other status line plugins
• Programmatic access to token counts
• Virtual environment management

• Setup issues: :checkhealth token-count
• Dependencies: :TokenCountVenvStatus
• Python not found: Ensure Python 3.7+ is in your PATH

MIT License