Privacy

When you paste a prompt and click Analyze, all pattern detection happens in your browser. The text is not sent to any server — not ours, not third parties. You can disconnect from the internet after loading the page and the analysis continues to work.

The shareable link uses a base64 hash of the prompt in the URL. This hash is visible in your browser's address bar but is not transmitted to the server in normal GET requests (hashes, by HTTP design, stay on the client).

When you click Rewrite with AI, the flow is different: the correction instruction text is sent from your browser to our /api/rewrite endpoint, and from there to one of the free partner AIs (Gemini, Mistral Small, or Llama 3.3 via Groq). The provider is selected automatically by the rotator, prioritizing whoever has available quota in the current window.

What Whet stores: only numerical performance metrics — specifically {providerId, scoreBefore, scoreAfter, delta, elapsedMs, success, timestamp}. These feed the live ranking at /whet-benchmark?tab=live. Your prompt text is never persisted — not the original prompt, not the correction instruction, not the rewritten result. The request passes through server memory, is answered, and the content is discarded. Only score numbers survive, and it is not possible to reconstruct the prompt from them.

What providers store: each provider has its own policy.

• Google Gemini (free tier) — the current AI Studio policy allows inputs and outputs to be used for model improvement. If this is a concern for your use case, avoid using automatic rewriting for sensitive prompts.
• Mistral (free tier) — the "Experiment" tier has its own clauses; see mistral.ai/terms.
• Groq (Llama 3.3 70B) — Groq operates the inference; see groq.com/privacy-policy.

Automatic rewriting is optional. If you prefer not to send text to third parties, the classic flow is always available: expand the "View correction instruction" section in the rewrite panel, copy the text, and paste it into your AI of choice — including local models like Llama running on your machine.

To prevent abuse of free tiers, there is a rewrite limit per IP per hour. Once the limit is reached, the panel shows the correction instruction and you can continue with the classic flow. The limit is based solely on the IP header of the request — no cookies, no sessions, no persistent identifiers.

Prompts above 4,000 characters are rejected by the rewrite route. Local analysis has no such limit.

• Whet uses Vercel Analytics — aggregated page view collection, cookieless, no individual tracking, no cross-site identification. Data is used solely to understand general product usage (access volume, pages visited). See vercel.com/docs/analytics/privacy-policy.
• We do not track individual users, use tracking cookies, or share data with ad networks.
• We do not store prompts, rewrites, or diagnostics in any database.
• We do not sell, share, or license data to third parties.
• We do not require signup, login, or any identification.

Questions about privacy? Contact us: hello@trywhet.com