Is AI private?

Tessera is a personal AI assistant with a memory you own. It is not an AI model itself: the thinking is done by the major AI models, and you pick which one answers. The memory belongs to you rather than to a model.

The short version

• Your conversations do not train AI. Tessera never uses your conversations or your memory to train anything of our own, and your questions reach the AI models on business tiers where conversations are not used to train the models by default.
• The AI company gets the context, never the person. Your question reaches the AI model with no account behind it, no email address, and nothing that builds up on their side. They read what is needed to answer, and have nowhere to file it.
• We have no ads business. Subscriptions are the whole business model, so there is no ad product your data could be sold to power.
• Nothing about you is a black box. Your memory opens like a settings page. You see everything that is known, and you can read every line, change it, add to it, or erase any of it at any time. Nobody at Tessera reads it to run the product, and the most sensitive parts are scrambled with a key belonging to your account alone, so our own view of the database shows nothing readable.
• You can take it or erase it at any time. Request a full copy of your data, delete a single fact, or delete the whole account. Deletion is real deletion.

Does AI train on my data?

Not through Tessera. We never use your conversations or your memory to train anything of our own.

When your question goes to an AI model, it goes under business terms rather than the consumer version of that product. That distinction matters more than most people realise. In the consumer apps, what you type can become training data by default. On the business tiers we use, the default is that it does not.

This is the question people ask most often, and it deserves a direct answer rather than a paragraph of hedging: your words are used to answer you, and they are not used to teach anything.

One honest footnote, because “not trained on” and “not kept” are two different promises and it is easy to blur them. Most of the models we send questions to keep nothing at all afterwards. One of them keeps a record of activity for up to thirty days so it can catch people abusing its service, and then deletes it. We would rather tell you that than round it off. The full privacy policy sets out exactly which is which.

A second footnote for the same reason. To find the right facts to bring to a question, Tessera turns your saved facts into a mathematical index, and that step runs through a specialist provider rather than on our own hardware. Their standard terms let them learn from what they are sent, so on August 1, 2026 we switched that off for our account: what we send them now is deleted as soon as it is processed and is not used for training. That switch does not reach backwards, so facts indexed before that date stay under their earlier terms. We would rather name that here than let “nothing is used for training” cover something it does not.

Who can see what Tessera remembers?

You can, completely. Your memory is not a black box: it opens like a settings page, one fact at a time, and you can read every line of it.

When you ask a question, the relevant pieces of that memory travel with the question to the AI model answering it, under the business terms above. Not all of it, every time. Just what is relevant to what you asked.

What does the AI company learn about me?

No account, and nothing that adds up over time.

This is the part most people assume works the other way around. When you use an AI app directly, everything you type is attached to your account there, kept on their side, and used to build a picture of you that you cannot open. Your name is on the file because your account is the file.

Through Tessera, your question arrives at the AI model with no account behind it, no email address, and nothing to connect it to. The model reads what it needs, answers, and has nowhere to file the answer. Nothing accumulates on their side from one question to the next.

Tessera holds the identity. The model does the thinking. Neither one has the whole picture, and only you do.

Worth being precise about what this covers. Before anything leaves Tessera, the names it knows are swapped for stand-ins. Your name, and the names of the people, pets and workplaces in your life, reach the AI model as labels such as “your sister” or “Coworker 1”. Email addresses, phone numbers, street addresses and ID numbers are taken out. Your birthday goes as your age, and your address as your city, or your state if you live somewhere small. The answer comes back with the real names put back in, so you never see the labels.

Four things this does not cover, stated plainly. When you talk to Tessera by voice, the transcription service hears the recording, names included. If you share a photo, the AI sees the whole photo. If you ask for a picture with a name in it, that name goes to the image model so it can draw it. And a name Tessera has never seen before can go out as written the first time. What never goes, in any of these, is the account it belongs to.

What can I check for myself?

All of it. Every fact Tessera holds is listed in the app, in plain language, grouped the way you would expect. You can read the whole thing in a sitting.

If something is wrong, edit it. If you never want it used again, delete it, and it stops travelling with your questions from that moment. Most products ask you to trust a policy page. This is the part that lets you check instead.

Can I delete what AI knows about me?

Yes, at two different sizes:

• One fact. Find it in your memory and delete it. It stops travelling with your questions immediately.
• The whole account. Delete the account and the data goes with it.

Deletion means the record is gone, not moved into an archive you cannot see. You can also request a full copy of your data at any time, and we send it to you.

Why a memory layer is more private than memory inside one AI app

When memory lives inside a single AI product, you generally cannot see all of it, cannot move it, and cannot take it with you when you leave. It is a feature of their product, held on their side.

Tessera keeps that memory on your side of the conversation. It is visible, editable, portable across AI models, and deletable, and it is the same memory no matter which model you are talking to today.

What we collect, and why

What you choose to tell Tessera about yourself, the conversations you have through it, and the ordinary account details needed to run a subscription. If you connect a calendar, the events you allow it to read. An invitation carries the names and email addresses of the people on it, so those arrive with the event and are stored with it, encrypted, and shown to you on the entry. We do not look those people up, build records about them, or contact them. That is the working set, and it exists so that answers fit your life instead of being written for nobody in particular.

We do not buy data about you from anyone else to fill in gaps.

Questions we get asked