I'm 12 and I built a local LLM gateway that tracks every token without sending data to the cloud.

• 1/9
  Most LLM cost trackers work the same way: send all your prompts to their cloud, they analyze them, and charge you $20-200/mo for the privilege.
  
  I didn't want to trust a third party with every message I send to ChatGPT, Claude, or my local models.
  
  So I built Token Guard. A local-only API gateway that runs in your system tray, proxies your LLM requests, logs every token to SQLite, and shows you a dashboard — all on your own machine. Zero telemetry. Zero cloud.
  
  https://tokenguard.pages.dev
  
  2/9
  How it works:
  
  1. Point your apps (OpenAI SDK, curl, anything) → http://localhost:8080
  2. Token Guard forwards to any provider — OpenAI, Anthropic, Google, OpenRouter, Ollama, vLLM
  3. Every request is logged: tokens used, cost, latency, model
  4. See it all in a clean dashboard
  
  That's it. No setup script. No Docker. No cloud dependency.
  
  3/9
  The format conversion is the magic part.
  
  You can use the OpenAI SDK to call Claude. Or the Anthropic SDK to call Gemini. Or any combination of the 3 × 3 matrix. Token Guard converts requests and responses between OpenAI, Anthropic, and Google formats automatically.
  
  Vision, tools, streaming — all converted on the fly.
  
  4/9
  Built with:

• Tauri + Rust (2MB binary, <100MB RAM)

• React + Tailwind CSS for the UI

• SQLite for local storage

OS keychain for API keys (they never touch disk)

Desktop app for Windows, macOS, and Linux. One binary, no runtime.

5/9
Pricing:

• Core: free and open source (Apache 2.0)

• Pro license ($4.99 one-time, lifetime): removes the support banner, enables auto-updates
  
  No subscription. No upsells. No data collection. Just a one-time "thanks for supporting the project" button.

  6/9
  Who's this for?

• Anyone using LLMs through an API and wondering where the money goes

• Self-hosting enthusiasts who keep everything local

• Developers who want one dashboard across every provider

• People tired of cloud monitoring tools that cost more than the API calls themselves
  
  7/9
  I'm 12 years old. I built this solo — Rust backend, React frontend, Tauri desktop shell, Cloudflare Workers for licensing, Paddle for payments.
  
  I started because I couldn't find a cost tracker that didn't send my data to someone else's server. So I built one.
  
  8/9
  It's not perfect. Provider presets could be smoother. The Google Gemini auth needs a fix. But the core loop works: set up the proxy → call any LLM → see instant cost tracking.
  
  If you try it and something breaks, tell me. I ship fixes fast.
  
  9/9
  Download: https://github.com/QQSHI13/tokenguard/releases
  GitHub: https://github.com/QQSHI13/tokenguard
  
  No cloud. No telemetry. Just your tokens, tracked locally.