Privacy policy

What we collect

• Account details. Your email address, your name, a scrypt hash of your password, when you created the account, and when you confirmed your email. If you join the waitlist while sign-ups are closed, we keep the email address you leave.
• Your style guide. The habits, examples, and reply preferences in your style guide, every earlier version of it, and for each change the time and whether it came from the editor or from a connected app.
• Marks. Text you mark as an example of a habit, and the rewrite you prefer, whether you mark it in the editor or through a connected app.
• Catalog suggestions. When you agree to send a mark to the shared catalog, we store the marked text, your rewrite, your critique, where the text appeared, the habit you chose or proposed, and when you agreed. We also store the name and version that the app reported for itself and the model that wrote the text. The app or you report the model; we do not check it.
• Connected apps. For each app you allow, such as Claude or ChatGPT: the name and return address the app registered with, the access you granted, when you connected it, when it last used your style guide, and the name and version the app reports when it connects. We store hashes of the app’s access tokens and refresh tokens, and never the tokens themselves.
• API tokens. The name you give each personal token, its first characters, a hash of the token, when you created it and last used it, and the name and version of the app that last used it.
• Rate-limit records. To stop abuse, we count sign-in attempts, sign-ups, password resets, and API and OAuth requests. Each record holds a time and a one-way hash of an IP address or email address, or an account, token, or app identifier.
• Sign-in cookie and theme. A cookie keeps you signed in, and we store only a hash of its value. A second cookie and your browser’s local storage remember whether you chose the light or dark theme. We use no advertising or analytics cookies.
• Logs. Our hosting provider records each request, including the IP address, the browser’s user agent, and the page or endpoint. Our server logs show only the domain of an email address.
• Text you compare. If you use the Eval page, we send the prompt you type to an AI model and show you the replies. We do not store the prompt or the replies.
• Text you check. When a connected app uses the check_text tool, or a hook script calls the check endpoint, we send the text to AI models through OpenRouter to check it against your style guide. We keep the text and the result only in server memory, for up to 10 minutes, so that the same check repeated in that time returns the same result. Text from apps whose terms limit how we use their data, such as Muse, skips that cache. We do not save checked text in our database or write it to our logs.

How we use it

• To run your account, sign you in, and send you email confirmation and password reset links.
• To keep your style guide and its history, and to give it to the apps you connect.
• To review catalog suggestions. Admins read each suggestion, and the text of a suggestion they accept can become part of the public catalog, without your name or email.
• To limit abuse with rate limits, and to keep the service secure and working.
• To answer you when you write to us.

We do not sell your personal data, use it for advertising, or use it to train AI models. We do not sell or license your text to the companies that make AI models. A model maker receives your text only when OpenRouter passes it to a model to serve one of the requests listed under Who else receives data.

Aggregate statistics

We may combine data from many accounts into statistics, such as how often each AI model shows each habit and how closely models follow style guides, and we may publish, share, or sell those statistics. They are de-identified and aggregated: they do not identify any person, and they contain no text that you or any other user wrote.

Research

Preferent has no research program today. If we offer one later, it will be opt-in, and we will ask for your explicit consent before any of your data goes into it. Data we collected before you opted in stays out unless you choose to include it, and you can withdraw at any time.

What connected apps receive

When you connect an app, such as Claude, ChatGPT, Cursor, or another MCP client, and choose Allow, that app can read your whole style guide. With write access, it can also add marks, add or change habits, change your reply preferences, and send suggestions to the catalog after you agree to each one. The app’s maker handles what the app receives under its own privacy policy. You can disconnect any app on your account page, and its tokens stop working at once.

Some apps, such as Meta’s Muse, limit by their own terms how we may use the data we receive through their connector. We use data from those apps only to answer your requests and to run, secure, support, and improve Preferent for you. We share it with a service provider only when a request you make needs it, we never use it to train AI models, and we leave it out of aggregate statistics and any research program. When you disconnect such an app or ask us to delete its data, we delete the data we received through it.

Who else receives data

We share personal data only with these service providers, and only for the work listed:

• Vercel hosts the website and the MCP server and keeps request logs.
• Neon hosts the Postgres database that holds your account, your style guide, and every other record in this policy.
• Resend sends our email, so it receives your email address and the content of each email we send you.
• OpenRouter passes text to the AI model that serves a request. It receives the text of catalog suggestions, without your name or email, when a model sorts them into catalog habits, the prompts you type on the Eval page, and the text you check with check_text or the check endpoint, along with your style guide. OpenRouter passes the text to the company that runs the chosen model, such as OpenAI, Anthropic, or Google. For checked text, we ask OpenRouter to use only model providers that do not keep the text or train on it.

Preferent admins can see the list of accounts with names and email addresses, and read catalog suggestions with the email of the person who sent each one. We may also disclose data when the law requires it, or to protect the safety of people or the service.

How long we keep it

• Your account, style guide, history, marks, connected apps, and API tokens stay until you delete your account.
• Sign-in sessions end after 14 days. App access tokens expire after an hour and refresh tokens after 90 days. We delete expired app tokens when the app next receives new ones.
• App registrations that never receive your permission are deleted after a day.
• Checked text and its result stay in server memory for up to 10 minutes.
• Rate-limit checks use the past hour of records, or the past 24 hours for text checks. We delete older records the next time the same limit is checked.
• Our hosting provider keeps request logs for a limited period under its own retention settings.
• Catalog suggestions stay after you delete your account, without the link to your account, so the review record and any catalog entry made from them remain. Write to us if you want a suggestion removed.
• Deleted data can remain in our database provider’s restore history for up to 30 days before it is gone.

Your controls

• On your account page, you can download your data as a JSON file: your account details, your style guide with its history, your catalog suggestions, your connected apps, and your API tokens without their secrets.
• On the same page, you can delete your account. Deletion removes your account, sessions, style guide and its history, API tokens, and every connected app with its tokens. Your catalog suggestions lose the link to your account.
• You can disconnect any app and revoke any API token at any time.
• You can edit any habit, mark, or preference in your style guide, and reset your password from the sign-in page.

Depending on where you live, you may have the right to access, correct, delete, or move your data, and to object to how we use it. Write to support@preferent.ai for any request this page does not cover. We answer within 30 days.

Children

Preferent is not for children under 13, and we do not knowingly collect data from them. People from 13 to 17 can use Preferent with a parent’s or guardian’s permission. If you believe a child under 13 created an account, write to support@preferent.ai and we will delete it.

Security

Every connection to Preferent uses HTTPS. We store passwords, session cookies, API tokens, and app tokens only as hashes, and email links work once and expire. To report a security problem, write to security@preferent.ai.

Where data is processed

Our service providers process data in the United States and in other countries where they operate. When data leaves your country, we rely on the safeguards those providers offer for such transfers.

Changes to this policy

When we change this policy, we update the date at the top of the page. If a change affects how we use data we already hold, we tell you by email or on the site before it takes effect.

Contact

Preferent answers privacy questions at support@preferent.ai and security reports at security@preferent.ai. The terms of service cover the rest of your use of Preferent.