Last updated October 24, 2026
Nyxie Focus uses AI to read your email so you don’t have to. That’s a lot of trust, so here is exactly how we protect it, in plain words.
Nyxie’s AI is Claude, made by Anthropic. Under Anthropic’s business terms it may not train on your email: it gets only the text Nyxie needs to write the summary, and Anthropic deletes it within 30 days. If Claude is ever unavailable, Cloudflare Workers AI, on the same network that runs Nyxie, writes the summary instead. Your email isn’t sent to ChatGPT, Gemini or any other AI service.
Nyxie reads each new email once to decide where it belongs and write a short summary. Looking back at, or filing, mail from before you joined uses only who sent it and its subject line, with no AI, except for the few emails it keeps for you, which are read like new mail. The email itself and its attachments are never stored. The short record of how it was sorted is erased after 30 days.
Your email is never used to train AI models, by us or anyone else, and we never sell or share your data or show you ads.
The AI can only decide whether an email needs your reply or is important to see, and add a “looks suspicious” warning when an email looks like a fake sign-in request (it can never remove a warning or send mail to Junk). It can’t send, delete, move or hide anything by itself. Nothing is ever sent without your tap, and Nyxie never deletes your mail. You can delete an email yourself: it goes to your mail app’s Trash, never erased for good, and Undo brings it back. Every move can be undone.
Reply and forward drafts. When you tap Reply or Forward, the AI writes a draft in your voice for you to read and change, and the buttons under it (Shorter, Warmer, Say yes…) only rewrite that draft. Nothing is sent until you tap Send, and only once you’re live. Drafts aren’t saved anywhere, never include account or card numbers, codes or passwords, and there’s no draft for an email flagged as suspicious or dangerous.
Questions about an email. When you ask Nyxie something while reading an email (“when does the rebate come?”), the AI reads that one email to answer. It only answers: it can’t act on the email, nothing is saved, and it won’t answer questions about mail flagged as suspicious or dangerous.
Learning from your sent mail. When you sign up, or ask Nyxie to learn again, the AI reads the subject lines of your recent sent mail (never the messages) and suggests a few topics and the kinds of mail you deal with. Nyxie keeps only those short labels, you choose what to keep, and they can only bring mail to your attention; your “never show me” list is your own choice, and it never touches people you know or security alerts.
Scammers sometimes hide instructions inside emails, hoping an AI will follow them. Nyxie treats every email as untrusted: the AI is told to ignore anything an email says to do, and even then it can only choose from a short list of safe actions that Nyxie’s own code checks. Sender names are treated the same way, and Ask Nyxie only makes a rule for a sender you actually mention. No request, from you or hidden in an email, can turn off the security checks: suspicious mail is always flagged, dangerous mail always goes to Junk, and your own rules can’t override either. Every change shows in Settings → Activity.
Summaries and previews hide account and card numbers (showing only the last 4 digits, the way banks do), one-time codes, passwords and Social Security numbers. Inside the app they show amounts and balances, so you can see what a bill or statement says at a glance. Phone notifications, Siri and spoken briefings hide every amount and balance, because a lock screen or a speaker can reach whoever is nearby. Open the email to see everything. One exception: when a sign-in or verification code arrives, the app shows it big with a Copy button for 15 minutes, read straight from the email each time and never saved. Never for suspicious or dangerous mail, and never in notifications, Siri or spoken briefings.
Due soon lists deadlines and meetings the AI found, but only from mail that reached Needs reply or Important: never from suspicious or quarantined mail, and never sale countdowns from promotions or newsletters. It shows no money amounts, and you can remove any item with “Not a deadline” or switch it off in Settings.
Your calendar is shown as it is, never written by the AI. Invites are answered only when you tap Accept, Maybe or Decline (Ask Nyxie can’t answer them), and there are no answer buttons on invites from senders you blocked or that Nyxie flagged as suspicious. Nyxie never creates, moves or deletes an event.
When Nyxie checks a link or shows you a preview, it happens on Cloudflare’s servers, far from your computer or phone. Nothing is clicked from your device, no forms are filled in, and nothing is downloaded.
Nyxie’s team sees account and billing details only, never your email. The one exception is a security vote: when you tap “This is safe” or Release on a flagged email, we see the sender’s domain and why it was flagged (never the email), so we can make Nyxie’s checks better for everyone. On a Business plan, whoever pays sees who’s on it and when they were last active, never their mail. Your Microsoft sign-in is stored encrypted.
Nyxie starts in preview mode and changes nothing in your mailbox until you turn on Go live. You can pause Nyxie any time, delete everything with one button in Settings, and remove its access from your Microsoft account whenever you like.
Questions? Write to privacy@nyxie.io. The full details are in our privacy policy.
