tcnt

TokenCount CLI

A Node.js command-line tool to count LLM tokens in text files. It can process local files or read from stdin.

Features

• Counts tokens using OpenAI's tiktoken library by default.
• Supports input from file path or stdin.
• Basic framework for adding other tokenizers.
• Handles text files only.

Installation

1. Clone this repository (or ensure you have the tokencount.js and package.json files).
2. Navigate to the project directory in your terminal.
3. Install dependencies:

   npm install

4. Make the script executable:

   chmod +x tokencount.js

5. Link the package to make the tokencount command available globally:

   npm link

Usage

Count tokens in a file:

tokencount /path/to/your/file.txt

Count tokens from piped input:

cat /path/to/your/file.txt | tokencount

Specify a tokenizer (default is openai-tiktoken):

tokencount --tokenizer openai-tiktoken /path/to/your/file.txt

Currently, openai-tiktoken is the primary supported tokenizer. A placeholder for gemini-text exists but will use openai-tiktoken as a fallback with a warning, as on-device Gemini tokenization is not yet implemented.

Get help:

tokencount --help

Supported Tokenizers

• openai-tiktoken: Uses the gpt2 encoding from OpenAI's tiktoken library.
• gemini-text (Placeholder): Currently falls back to openai-tiktoken. On-device support for Gemini tokenization is a future consideration pending available libraries.

Limitations

• Text Files Only: This tool is designed for text files. Attempting to process binary files will result in an error or incorrect counts.
• On-Device Tokenization for Gemini: True on-device tokenization for Gemini models is not yet implemented.

Development

To contribute or modify:

• The project uses ES Module syntax (import/export).
• The main script is tokencount.js.
• Tokenizer logic is handled within the .action(...) callback in tokencount.js.
• To add a new tokenizer, you would typically:
  1. Install any necessary Node.js package for that tokenizer.
  2. Import necessary functions from the package using import.
  3. Add a new else if condition for your tokenizer's name in tokencount.js.
  4. Implement the token counting logic within that block.
  5. Update this README and the help messages.