01What the website chat does
The "Ask Tractionora" chat is an AI assistant, not a person. It answers questions about Tractionora from approved published information. It cannot see anyone's account, change anything, save an enquiry or send a message to our team. It may offer links to Start Free, Request Help, Join Early Interest or the contact form.
02What is sent to TypeSafe and OpenAI, and why
When a question needs AI, our server works in up to two steps. TypeSafe makes the first decision: what the question is about and which approved facts fit. It receives your message, up to four recent earlier messages from the same chat and a short note of your stated goal and business. If TypeSafe is unavailable, OpenAI makes that decision instead. For some questions OpenAI then writes the answer, and it receives your message, up to eight recent earlier messages from the same chat (so follow-up questions make sense) and the approved Tractionora facts the answer must stay within.
Not sent from the chat: anything from the contact or enquiry forms, such as the name and email address you enter there, cookies, or account information. The chat never receives those form details. Personal details you choose to type into the chat, such as your name or an email address, are part of your message and are included in what is sent to TypeSafe and, where it writes the answer, OpenAI, so please keep private details out of chat. We send no IP address or visitor identifier; TypeSafe and OpenAI receive the request from our servers. Some common questions and sensitive topics (complaints, security, account problems) are answered with fixed text or a pointer to a person, without an AI provider. Messages that look like passwords or keys are refused before anything is sent. The automatic acknowledgement to a contact form is separate and is described in section 07.
03What Tractionora stores
We do not store chat transcripts. The conversation stays in your browser page while the chat is open and is not saved to your device or our database. Our logs record only a technical event, a status and the provider's request reference. Our hosting provider keeps its own technical records. Chat is separate from enquiries: nothing you type in chat is copied into an enquiry.
04What TypeSafe and OpenAI do with it
TypeSafe is a separate company with its own terms. Its published terms say it will not use customer data to train its AI models without the customer's consent, and that it may keep technical records derived from use, such as logs and classifications, and use customer data to detect fraud and abuse. We make no statement about where TypeSafe processes chat or contact messages, how long it keeps them, or whether Zero Data Retention applies to us.
OpenAI is a separate company with its own rules. For our website project:
- Training. Data sent through OpenAI's API is not used to train its models unless the customer opts in. Our organisation has OpenAI data sharing switched off.
- Retention. OpenAI keeps abuse-monitoring records of API use for up to 30 days by default, or longer if the law requires. Our website project uses this standard setting.
- Where. Our project uses OpenAI's global data setting. We do not say that your messages are stored or processed in any particular country or region.
- Zero Data Retention. We do not have Zero Data Retention, and do not claim it.
- Stored answers. When OpenAI writes an answer we ask it not to keep the response as stored data. TypeSafe is the primary decision provider. When OpenAI makes a fallback decision about what the question is about, because TypeSafe is unavailable, we use OpenAI's default settings; we have not confirmed whether OpenAI keeps anything beyond the abuse-monitoring records for that step, and we do not claim it does not.
Because of this, we cannot promise to delete a chat or contact message from TypeSafe's or OpenAI's records.
05What not to share
Please do not type passwords, sign-in codes, API keys, payment or bank details, government ID numbers, health information or Google access details into chat. We try to refuse obvious cases but cannot catch everything. A message that needs AI and passes is sent to TypeSafe and, when needed for a fallback decision or to write the answer, to OpenAI, and may be held as described above. Questions answered by our fixed rules need not reach either provider. For account help, use the contact page and never include credentials.
06Mistakes and human help
AI answers can be wrong or out of date. Check prices and plan details on the pricing page. A chat answer is not a commitment, a contract or professional advice. A person can help at any time: use the contact form or the Request Help form. The chat says plainly when nothing has been sent to the team, and never claims an enquiry is saved unless you completed the form and confirmed your email.
07Contact-form acknowledgements
When you send a contact message we email our team and may send you an automatic acknowledgement. The website drafts that acknowledgement with AI in up to two steps. First, TypeSafe receives the topic you chose and your message text (up to 5,000 characters) and decides which approved answer fits. It does not receive your name or email address. If TypeSafe is unavailable, OpenAI makes that decision instead. Only when a message needs a more specific answer does OpenAI write it, receiving your name, topic and message text with approved facts, but not your email address. We ask OpenAI not to keep the written response, and the abuse-monitoring rules in section 04 still apply. Emails we send are also held by Resend for 30 days on our current plan. Messages about security, complaints and account problems receive fixed wording, and a person reads your message and replies. Please do not put passwords, payment details or health information in a contact message. The enquiry forms (Request Help, Join Early Interest) do not use AI.
08AI inside the Tractionora application
The website chat knows nothing about you. Inside the application, AI may help with a customer's own workspace on that customer's instructions and within its permissions. No model output grants authority to make a change: nothing is published until you approve the exact version you reviewed.
09Questions
Email legal@tractionora.com. See also the privacy policy and the subprocessor list.