tōks

tōks ("tokes" as in "tokens") counts tokens for files and directories across LLM providers. See how much of a model's context window your content will consume before you spend the tokens. This is particularly useful for context engineering & token budgeting when building software using AI agents.

Installation

Install as a tool with uv:

uv tool install toks

OR, if you're still a "pipper"

pip install toks

The ct ("count tokens") shorthand alias is also available — ct src/ is equivalent to toks src/.

Quick Start

1. Configure your providers

toks setup

The interactive wizard walks you through selecting providers, entering API keys, and choosing models.

2. Count tokens

# Single file
toks src/main.py --for claude

# Entire directory with glob filter
toks src/ --for claude --glob "*.py"

# Quiet mode (just the number)
toks README.md --for claude -q

# Pipe from stdin
cat file.py | toks - --for claude

Output

Provider: claude (claude-opus-4-6)

File                   Tokens  Agent    Web    API
toks/                [17,911]   1.8%   9.0%   1.8%
├── providers/        [4,174]   0.4%   2.1%   0.4%
│   ├── __init__.py       324  <0.1%   0.2%  <0.1%
│   ├── base.py           474  <0.1%   0.2%  <0.1%
│   ├── claude.py         600  <0.1%   0.3%  <0.1%
│   ├── gemini.py         643  <0.1%   0.3%  <0.1%
│   ├── grok.py         1,315   0.1%   0.7%   0.1%
│   └── openai.py         818  <0.1%   0.4%  <0.1%
├── __init__.py            33  <0.1%  <0.1%  <0.1%
├── __main__.py            28  <0.1%  <0.1%  <0.1%
├── cli.py              3,188   0.3%   1.6%   0.3%
├── config.py           1,338   0.1%   0.7%   0.1%
├── output.py           2,260   0.2%   1.1%   0.2%
├── registry.py           978  <0.1%   0.5%  <0.1%
├── runner.py           1,146   0.1%   0.6%   0.1%
├── scanner.py          1,318   0.1%   0.7%   0.1%
└── setup.py            3,448   0.3%   1.7%   0.3%

Total: 17,911 tokens
  Agent (1M):  1.8%
  Web (200K):    9.0%
  API (1M):    1.8%

Three percentage columns show context window usage for each provider's interfaces: coding agent, web chat, and API. The Agent and Web columns populate when you configure your plan and coding agent model via toks setup.

Supported Providers

All token counting is done via each provider's REST API. No local tokenizer libraries are needed.

File Handling

When scanning directories, toks respects .gitignore rules by default — files matching gitignore patterns are skipped. Nested .gitignore files are also respected. Use --no-gitignore to include all files regardless. Binary files are excluded by default (--include-binary to override), and files over 50MB are skipped (--max-size to adjust).

Configuration

Config files live in ~/.config/toks/:

• config.toml -- provider settings, default model, plan tier
• .env -- API keys

Re-run toks setup at any time to update your configuration.

Commands

toks <target> [options]    # Count tokens (default command)
toks setup                 # Interactive configuration wizard
toks models --refresh      # List known models / refresh registry

ct can be used anywhere in place of toks.

Options

Documentation

• Technical Documentation -- architecture, module details, tutorials
• Feature Specification -- complete design specification

License

MIT