Where your words go

Emes Assistant runs on a computer you own. Your appointments, notes, lists and everything it learns about how you work are kept in a file on that machine. There is no account, no server of ours holding your data, and nothing for us to hand over, sell or lose — because we never have it.

Some things do leave your machine, and you should know exactly what they are: the words you say to it go to Anthropic to be understood, your notifications go through a push service to reach your phone, and — if you connect them — your calendar (and, if you turn it on, your Gmail) is read from Google, your coursework from Canvas and your mail from Microsoft 365. Each is described below.

On your computer, in a database file you can open, copy or delete: appointments and reminders, your daily lists, links you saved to read, rules it has learned about how you like things done, phrasings it has learned from you, a record of notifications it sent, and your settings.

Audio is never stored. When you speak to it, your phone turns that into text and the audio is discarded. The text is used to work out what you meant and is then dropped as well — it is not written down. The one exception is described under “Testing” below, and it is switched off by default.

Anthropic. Every instruction you give it is sent to Anthropic’s Claude to be turned into something the software understands — “dentist Thursday at 3” into an appointment. This is how it works at all. Anthropic receives the words you say to it, along with the parts of your schedule needed to answer. It reaches Claude one of two ways: the Claude API, or — on a machine where its owner chose this — the Claude subscription of the person signed in to Claude on that machine, in which case that account’s terms with Anthropic apply. Either way, a web search it runs to answer you happens on Anthropic’s side. Their handling is governed by their own terms. There is a setting that uses a model running on your own machine instead, and then nothing you say leaves the house.

Push notifications. Reminders reach your phone through a push service: Pushover (pushover.net) on a machine set up with it, and otherwise the public server at ntfy.sh. The text of a notification passes through that service, which means the title of an appointment can be visible to it — “Dentist at 3 — Dr Weiss”. For an alarm that rings until you answer it, Pushover also tells the software when you tapped Acknowledge. This is the weakest point in the design and we are not going to pretend otherwise. It can be pointed at a server you run instead, and we are working on making the notifications carry no detail at all.

Google, only if you connect it. If you link a Google account, the assistant reads your calendar so that what is already in your diary shows up alongside what you tell it, and reads your YouTube subscriptions if you use the morning video. If you turn it on, it also reads Google Sheets you name that are calendars — a club that keeps its term in a spreadsheet — once a week or when you ask, and books what is in them on your own machine (updated 22 September 2026). All of it is read-only: it cannot change or delete anything in your Google account. Nothing is ever written back, and your Google data is not sent anywhere except between your machine and Google, and to Anthropic for the parts needed to answer you (see above).

Your Gmail, only if you turn mail reading on. It is off unless you chose it when signing in with Google (updated 16 September 2026). When it is on, your own machine reads your Gmail on your behalf, read-only, when you ask it something about your mail — searching it, and opening a message to read its subject, sender, date and text. It cannot send, delete, label or change anything, and it never opens attachments; it only sees their names. No message content is kept: nothing from your mail is written to the database or to any log. The only thing that stays is what you agree to — a task or a deadline you said yes to, in the words you approved. To answer you, the parts of a message it read are sent to Anthropic in the same way as everything else you ask it (see above), and if you asked through another chat app, its answer goes back through that app. The one exception to keeping nothing is the testing record described below: if you agreed to it, the assistant’s replies are part of it, so a reply that quoted your mail is kept there until that record erases itself. Your mail is not used to train any AI model, and is never sold or used for advertising. Emes Assistant’s use and transfer to any other app of information received from Google APIs will adhere to the Google API Services User Data Policy, including the Limited Use requirements. Turning it off: sign in with Google again without mail, or remove the app at myaccount.google.com/permissions.

Canvas, only if you connect it. With an access token you create in your school’s Canvas, the assistant reads your courses, assignment due dates, whether each was handed in, and your course grades. It only reads; it never submits or changes anything. When you ask it to, due dates are copied onto your machine as deadlines (course, assignment name, date, and whether it is finished). Grades are read when you ask and are not kept. What it reads to answer you is sent to Anthropic like the rest of your schedule. Other course sites such as Packback, Pearson and Norton are not read at all; only what your course lists inside Canvas is seen.

Microsoft 365 mail, only if you connect it. If you sign in with a work or school Microsoft account, the assistant can search and read your mail through Microsoft Graph, with read-only permission. It cannot send, move or delete mail. What it reads is used to answer you in that moment and is not stored on your machine: no message, subject or sender is kept. What can be kept is a deadline or task you agree to add because of something in it. Like anything else it uses to answer you, the parts of a message it reads are sent to Anthropic, and to the app you are talking to it through if that app belongs to someone else. Your school may need to approve the connection first, and can withdraw it.

Links you save. When you save something to read later, your machine fetches that page once to get its title. That website sees a request from your home connection, as it would if you had opened it.

WhatsApp and iMessage notifications, only when you ask. Added 16 September 2026. On a machine with a Windows computer of its own, when you ask about your messages (“what did Zach say on WhatsApp?”), the assistant reads the message notifications Windows is already holding on that computer — WhatsApp Desktop, and iMessage and texts that Phone Link brings over from your iPhone. It reads what the notification shows: who or which chat, and the preview text. Not your chat history, and only what is still in the notification centre. Nothing reads them in the background, and none of them is kept: they are not written to its database, its logs, or the testing transcript. To answer you, the previews it read are sent to Anthropic in the same way your own words are, and the answer goes back through whichever app you asked in.

Your phone and your computer. The connection between them runs over a private network (Tailscale) which is encrypted end to end. We do not sit in the middle of it.

If we issued the key that lets your assistant reach Anthropic, we can see how much it was used and what it cost. We cannot see what you said, what is in your calendar, or anything in your lists. Your requests go from your machine to Anthropic directly; they do not pass through us.

If something breaks and you want help, you can send a diagnostic report — you will be shown exactly what it contains before it sends, and it is never taken without you choosing to send it.

The software normally keeps no record of what you say. During a limited testing period we may ask a small number of people to switch that off temporarily, so that we can see the cases where the assistant misunderstood them — the only reliable way to fix it.

If you are asked and you agree, then and only then:

• The words you type or speak, and what the assistant made of them, are saved on your own machine. Never audio.
• It is kept in its own file, separate from everything else, so erasing it removes it completely.
• It has an end date set before it starts. The software deletes it on that date by itself, and refuses to run without one.
• It says so the whole time — on the screen, in red, unmissably.
• You can erase it at any moment from your phone, and change your mind without giving a reason.
• One person reads it, only to fix how the assistant understands people. It is never shared, sold, or used to train anything.

This is off unless you have been asked and said yes. Saying no changes nothing about how the assistant works for you.

Your data is a file on your computer. Uninstalling the software and deleting that file removes everything, permanently, without asking anyone. There is nothing on our side to request the deletion of.

Disconnecting Google can be done from your Google account at any time, and takes effect immediately. Deleting the access token in Canvas disconnects Canvas, and removing the app from your Microsoft account (or your school removing it) disconnects mail.

Emes Assistant is not intended for anyone under 16 and we do not knowingly provide it to them.

If this page changes, the date at the top changes with it. If a change affects what leaves your machine, people using it will be told directly rather than left to notice.

Questions, or anything here that does not match what you are seeing: info@leonsurplusdeals.com.