⬡ OFFTOCO

OFFline TOken COunter

Count tokens for ChatGPT, Claude and Gemini simultaneously. Get a SHA-256 fingerprint of your text. 100% offline — your text never leaves your device, not even if attach files.

---

What is a token?

AI models don't read word by word — they break text into fragments called tokens (roughly 3–4 characters each). Every model has its own way of splitting text, so the same sentence gives different token counts on GPT, Claude and Gemini. Token counts matter because:

• Models have context limits measured in tokens
• API usage is billed per token
• Prompt design depends on knowing how many tokens you are using

Offtoco counts all three at once, instantly, with no internet connection required.

---

Screenshots (run locally or in hosting from low-end ones e.g., Namecheap)

Web app in any web browser and OS — light mode [LIVE DEMO ON GITHUB PAGES]

Web app in any web browser and OS — dark mode [LIVE DEMO ON GITHUB PAGES]

CLI output in terminal (Linux / macOS / Windows console)

Windows desktop popup after doing a right-click on any file...

... to show the next interface:

Download — no installation, no internet needed after download

All downloads are on the Releases page.

---

Web app — works on any computer or OS

Unzip once, open in browser, done. No server, no Node.js, no internet.

1. Download offtoco-web.zip
2. Unzip it — you get a folder called offtoco-web
3. Open offtoco-web/index.html in any browser (Chrome, Firefox, Edge, Safari)
4. Paste text, drop files, copy counts — everything works offline

To put it on your own server or intranet, copy the offtoco-web folder to any static file host — nginx, Apache, Caddy, GitHub Pages, Netlify, a USB stick served by Python. No backend required.

# Example: serve locally with Python (any OS)
cd offtoco-web
python3 -m http.server 8080
# Open http://localhost:8080

# Example: nginx on any Linux server
server {
    listen 80;
    server_name tokens.yourdomain.com;
    root /var/www/offtoco-web;
    index index.html;
}

---

Windows desktop app

Right-click any file in Explorer to count its tokens. No terminal needed.

1. Download Offtoco Setup 0.1.0.exe
2. Run the installer — pick your install folder, click Install
3. Register the right-click menus (one-time setup, no admin needed):

$exe = 'C:\Program Files\Offtoco\Offtoco.exe'
$cmd = "`"$exe`" --file `"%1`""
reg add "HKCU\Software\Classes\*\shell\Offtoco" /ve /d "Count Tokens (Offtoco)" /f
reg add "HKCU\Software\Classes\*\shell\Offtoco\command" /ve /d $cmd /f
New-Item -Path "HKCU:\Software\Classes\Directory\Background\shell\Offtoco" -Value "Count Clipboard Text (Offtoco)" -Force
New-Item -Path "HKCU:\Software\Classes\Directory\Background\shell\Offtoco\command" -Value "`"$exe`"" -Force

Now right-click any file in Explorer and choose Count Tokens (Offtoco).

To remove the right-click menus:

Remove-Item "HKCU:\Software\Classes\*\shell\Offtoco" -Recurse -Force -ErrorAction SilentlyContinue
Remove-Item "HKCU:\Software\Classes\Directory\Background\shell\Offtoco" -Recurse -Force -ErrorAction SilentlyContinue

Popup layout:

+- [hex] OFFTOCO --------------------------------- [x] -+
|  GPT     o200k_base         1,234   tok   [ Copy ]    |
|  Claude  Anthropic          1,189   tok   [ Copy ]    |
|  Gemini  256k vocab         1,301   tok   [ Copy ]    |
|  SHA-256 a3f4b2c1...f1a2            [ Copy ]          |
+- github.com/PacifAIst/Offtoco ----------- [ Close ] --+

---

CLI — Windows, Linux, macOS

Count tokens from the terminal or script it into your workflow.

Download and extract the zip for your platform:

Platform	Download	Extract and rename to
Windows	offtoco-cli-windows.zip	offtoco.exe
Linux	offtoco-cli-linux.zip	offtoco
macOS Intel	offtoco-cli-macos.zip	offtoco

Linux / macOS — make executable after extracting:

chmod +x offtoco
./offtoco hello world

Windows:

.\offtoco.exe hello world

All usage options:

offtoco hello world                  # text from arguments
offtoco -t "The quick brown fox"     # explicit text flag
offtoco -f my-document.txt           # read from file
cat file.txt | offtoco               # pipe from stdin
offtoco --json -f file.txt           # JSON output for scripting
offtoco --help                       # show all options

Output:

GPT              18  tokens
Claude           17  tokens
Gemini           22  tokens
SHA-256 c02d938580319068f8ab754ff664f289f13ab0bf4c5245acc29b14b668b5a9f8

JSON output (--json flag) for use in scripts:

{
  "gpt": 18,
  "claude": 17,
  "gemini": 22,
  "sha256": "c02d938580319068f8ab754ff664f289f13ab0bf4c5245acc29b14b668b5a9f8",
  "chars": 63
}

---

Tokenizers

Model	Vocabulary	Package	License
GPT (o200k_base)	200,000 tokens	js-tiktoken	MIT
Claude (Anthropic BPE)	64,739 tokens	@anthropic-ai/tokenizer vocab + js-tiktoken engine	Apache-2.0
Gemini	256,000 tokens	@lenml/tokenizer-gemini	Apache-2.0

Zero WASM. The standard Claude tokenizer package ships a WebAssembly binary that browsers reject. Offtoco extracts the raw vocabulary at install time and uses the same pure-JS engine as GPT — counts are bit-for-bit identical, with no WASM in the browser bundle.

---

Privacy

Network requests	None — ever
Telemetry	None
Analytics	None
Auto-update checks	None
Data stored	Nothing — not even locally
Verification	Full source code in this repository

The web app can be opened directly from a USB stick on an air-gapped machine. The CLI binary has no network code. The desktop app has no network code. All tokenizer vocabularies are bundled at install time.

---

For developers — build from source

Requirements

• Node.js >= 20 — nodejs.org
• pnpm >= 8 — npm install -g pnpm

Install

git clone https://github.com/PacifAIst/Offtoco.git
cd Offtoco
pnpm install
# postinstall auto-generates core/claude-vocab.js

Verify

pnpm run smoke
# GPT=18, Claude=17, Gemini=22 — OK

Web app

pnpm web:dev        # dev server -> http://localhost:5173
pnpm web:build      # production -> dist/web/

Create the web zip (unzip-and-run distribution)

New-Item -ItemType Directory -Force -Path dist\zips\offtoco-web | Out-Null
Copy-Item dist\web\* dist\zips\offtoco-web\ -Recurse
Compress-Archive -Path dist\zips\offtoco-web -DestinationPath dist\zips\offtoco-web.zip -Force
Remove-Item dist\zips\offtoco-web -Recurse -Force

CLI binaries (standalone, no Node required)

# Windows PowerShell — builds for Win, Linux, macOS
pnpm add -D @yao-pkg/pkg
npx esbuild cli/offtoco.js --bundle --platform=node --target=node20 --outfile=cli/offtoco.bundle.js --external:electron --format=cjs
npx pkg cli/offtoco.bundle.js --config cli/pkg.config.json --targets "node20-win-x64,node20-linux-x64,node20-macos-x64" --output "dist/cli/offtoco"

# Linux / macOS
bash cli/build.sh

Create zip for each binary:

New-Item -ItemType Directory -Force -Path dist\zips | Out-Null
Compress-Archive -Path dist\cli\offtoco-win.exe -DestinationPath dist\zips\offtoco-cli-windows.zip -Force
Compress-Archive -Path dist\cli\offtoco-linux   -DestinationPath dist\zips\offtoco-cli-linux.zip   -Force
Compress-Archive -Path dist\cli\offtoco-macos   -DestinationPath dist\zips\offtoco-cli-macos.zip   -Force

Windows desktop installer

cd desktop
npm install --legacy-peer-deps
npx esbuild main.js --bundle --platform=node --target=node20 --outfile=main.bundle.js --external:electron --format=cjs
npx electron-builder --win
# Output: desktop\dist\Offtoco Setup 0.1.0.exe

Repo structure

Offtoco/
|
+-- package.json              pnpm root (type: module)
+-- vite.config.js
+-- README.md
+-- docs/                     screenshots for this README
|
+-- core/                     shared tokenizer logic — Node and browser
|   +-- tokenizers.js         initTokenizers(), countAll() -> {gpt, claude, gemini}
|   +-- sha256.js             sha256hex() — SubtleCrypto / Node crypto
|   +-- format.js             fmt(1234567) -> "1,234,567"
|   +-- smoke-test.js         pnpm run smoke
|   \-- claude-vocab.js       auto-generated by postinstall, gitignored
|
+-- scripts/
|   \-- gen-claude-vocab.js   extracts claude.json -> plain ES module
|
+-- web/                      browser app (Vite 5)
|   +-- index.html
|   +-- main.js
|   \-- style.css
|
+-- cli/                      command-line tool
|   +-- offtoco.js            entry point
|   +-- pkg.config.json       asset manifest
|   +-- build.sh              Linux / macOS build
|   \-- build.ps1             Windows build
|
\-- desktop/                  Windows Electron app
    +-- main.js               main process
    +-- preload.js            IPC bridge
    +-- popup.html/css/js     popup UI
    +-- package.json          electron-builder config
    +-- install-registry.ps1
    +-- uninstall-registry.ps1
    \-- build.ps1             esbuild + electron-builder

---

Contributing

1. Fork and clone
2. pnpm install
3. Edit in core/, web/, cli/, or desktop/
4. pnpm run smoke must pass
5. Open a pull request

Hard constraints: zero network calls and zero WASM in the browser bundle.

---

License

GPL-3.0 (c) PacifAIst

Tokenizer vocabularies keep their upstream licenses (MIT and Apache-2.0), both compatible with GPL-3.0 downstream distribution.

---

Made with ❤️ for the Local AI Community by PacifAIst