What we do with what you tell us.

What we collect

What you give us, plus page statistics that use no cookies. We do not run fingerprinting or advertising trackers on this site, and Google Analytics runs only if you opt in.

• Contact form. Name, email, and the message body. Optionally: company, role, the kind of inquiry, a subject, project stage, budget, and timeline if you choose to share them. It reaches us as an email.
• Project planner. Your answers to the planner steps plus name, email, and (optionally) company and phone number. Your answers and the plan generated from them are emailed to us and to you, and saved in our database.
• AI readiness self-check. Your score is worked out in your browser, and nothing reaches us unless you ask for the full report. If you do: your answers and scores plus name, email, and (optionally) company are emailed to us and to you, and saved in our database.
• Voice / chat assistant. What you type or say to the assistant, and what it answers. Each exchange is saved in our database for six months, with the page you were on, any name, email or company you give the assistant, and a shortened hash of your IP address. If you ask the assistant to fill the contact or planner form, the same fields above are sent. A support request it drafts for you is sent only when you choose: by email or through our support form yourself, or with the card's Send button, which files it in our support system together with the email address you type.
• Support requests. If you are a client and ask for help through our support form at support.mentalbound.com, by email to support@mentalbound.com, or with the Send button on a request the assistant on this site drafted with you: your name if you give it, your email address, the site or project concerned, the subject and the message. From the form we also keep a keyed hash of your IP address, never the address itself. Each request becomes a ticket in our support system, together with our replies. If you use the form or the Send button, one confirmation email goes to the address you give, with the request number and your text. An AI model (Claude, by Anthropic) helps us sort requests and prepare replies. A person reads every request and sends every reply. When the assistant on this site hands you to the form, your draft travels in the page address, so it can also appear in our hosting provider's request logs and in your browser history.
• Page statistics (no cookies). Vercel Web Analytics counts page views without cookies and without storing anything on your device: the page address, the referring site, browser, operating system, device type, and approximate location. It tells visitors apart by a hash that is discarded after a day, so nobody is followed across days or across sites. It also counts a few events from the assistant — that it was opened, that a message was sent, what kind of action it took, that a support request was filed from it — never what was said.
• Google Analytics (optional). If you accept the cookie banner, Google Analytics records pages visited, referrer, device and browser type, and approximate location derived from IP. If you decline, it is not loaded at all.
• Server logs and rate limiting. Standard request logs (IP, user agent, path, timestamp) kept by our hosting provider for security and debugging. We also hold your IP in memory to rate-limit our API endpoints and to block addresses that probe the site for weaknesses — a minute at a time for rate limits, up to an hour and a half for a blocked address. The IP itself is never written to our database; a shortened hash of it is saved with assistant conversations and with planner and self-check submissions, so that entries from one visitor can be told apart.

Why we collect it

• To reply to you. The contact form, the planner and the self-check exist so we can answer, and assistant conversations are saved so that an enquiry made there is not lost when you close the tab. Support requests become tickets so that nothing you report is lost and we can see the history of your project.
• To deliver work. If we move into a project together, we keep what you sent so we can scope and follow up.
• To improve the site. Server logs help us catch errors and abuse; page statistics and opt-in analytics show us which pages actually help visitors; saved assistant conversations show us where its answers fall short.

Where the data goes

We use a small number of vetted processors. None of them sell your data.

• Resend delivers contact, planner and self-check emails to our inbox, and the copy of your plan or report to yours.
• Vercel hosts the site and our support system, stores request logs, and provides the cookie-free page statistics.
• Neon hosts the database that holds assistant conversations and planner and self-check submissions.
• Supabase hosts the database of our support system, in Frankfurt: the tickets, what you sent through the form or by email, and the replies we draft.
• Proton Mail is our mailbox. Mail you send to support@mentalbound.com arrives there, and our replies to support requests leave from there.
• Mailgun (EU region) carries incoming support mail from our mailbox into the support system.
• OpenAI transcribes your speech (Whisper) when you speak to the assistant and your browser has not already done so. Audio is sent for transcription only and is not retained for model training under our API configuration. Most browsers transcribe first with their own built-in speech recognition, which is run by the browser's maker (for example Google for Chrome, Apple for Safari) under its own terms.
• Anthropic powers the assistant's responses and the project planner's analysis, and helps us sort support requests and prepare replies to them. Anthropic may process that text outside the EU, under its data-processing agreement.
• Google provides analytics (Google Analytics 4) — only after you accept the cookie banner. Ads features are disabled; the data counts visits, it doesn't build advertising profiles.

Data sent to these providers is processed under their data-processing agreements. They do not retain your conversation content for model training in the configurations we use.