Token Wise

Estimate how long your prompt is before you send it, and rewrite it in one click to make it shorter and clearer.

Token Wise adds a small badge above the prompt box on the AI chat sites you already use. As you type, it estimates how long your prompt is and gives a simple low / medium / high guide, so you can see when a prompt has grown longer than it needs to be. Optimize in one click When a prompt looks long, click Optimize. Token Wise suggests a rewrite that is clearer, more specific and shorter, and shows the original and the rewrite side by side before you send. Choose the focus that suits you: Balanced, Shorter, Clearer or More specific. Keyboard shortcut Press Alt + Shift + O to rewrite the current prompt without leaving the keyboard. Where it works It runs on the major AI assistant sites and on multi-model chat platforms. The full, current list is shown in the extension settings and on our website, so what you see in the product always matches what is actually supported. Honest numbers All figures are estimates produced locally in your browser. Token Wise has no access to your provider account, billing, credits or quotas, and never shows a billed amount. Your prompts stay private Prompts stay in your browser unless you click Optimize. When you do, the text is sent over HTTPS to our rewriting service and is not stored. A local-only mode keeps everything offline. History and dashboard A side panel keeps your last 100 rewrites with export to CSV, and the popup shows daily, weekly and all-time activity. Everything is stored in your browser. Subscriptions are handled by Paddle as Merchant of Record. v1.2.4 — Bug fixes Fixed a failure where the Optimize button returned a generic error ("HTTP 400") in the workspace panel and on unsupported sites. Multiline prompts with Markdown formatting (headings, bullets, quotes, separator lines) are now handled correctly; prompts up to 30,000 characters are supported. Clearer error messages for over-length prompts, sign-in issues, and usage limits. Cost estimates no longer fluctuate when the selected model name is detected or lost — the estimate basis is now consistent. Reduced an overly eager "site layout changed" warning so it appears only when the page genuinely fails to settle, and never on sites the user has disabled. No changes to permissions, host access, or data handling.