See exactly what your
AI is costing you.

No. We monkey-patch fetch in your browser to read response metadata only — the usage object that platforms return alongside their answer (which contains numbers like input_tokens: 1247). The actual response text is never inspected, stored, or sent anywhere. The first 100 characters of your prompt are stored locally in browser storage so you can see it in your history — that storage never leaves your machine.

When the platform exposes the real usage object in its response (Anthropic and OpenAI both do for most accounts), we use that — 100% accurate. While you're typing, we run a fast client-side approximation that's within ~10-15% of the real BPE tokenizer. The overlay's dot turns bright teal for ~4 seconds when real numbers from the API arrive.

Because the data we need only exists at the moment you actually send a prompt — we have to be inside the page. A web app could only show what you've manually exported, which defeats the purpose. We're working on Firefox and Safari support next.

No measurable impact. The overlay updates on input events (which fire anyway). The network hook clones response streams for inspection, but the original stream the page reads is never blocked or modified.

Your Free features keep working forever (the counter, the daily summary). You'll lose the cross-platform dashboard, prompt coaching, and unlimited history — but your data stays in your browser, you don't lose anything you tracked. Resubscribe anytime and pick back up where you left off.

We're a passive observer of network traffic in your own browser, on a session you authenticated yourself. No automation, no scraping, no API key reuse. We're under counsel review for an explicit statement before public launch.

Working closed beta now. Public Chrome Web Store listing in the next 4-6 weeks. Waitlist members get access first and a launch discount on Pro.