Tessera
Explainer
What is AI memory?
The difference between being answered and being known.
A short, plain explanation of the term, what it is not, where it lives, and the questions worth asking before you let anything remember you.
AI memory is a standing record of who you are, kept outside any single conversation, that an AI can draw on when it answers you. Without it, every chat starts with a stranger. With it, the answer already accounts for your situation, your constraints and your preferences.
That is the whole concept. Everything below is detail.
What AI memory is not
It is not a bigger context window. A context window is how much text a model can hold in mind during one conversation: working attention, not a record. It empties when the conversation ends. A model with an enormous context window will still greet you tomorrow with no idea who you are.
It is not chat history. Scrolling back through old conversations is an archive you have to search. Memory is the part that shows up in the answer without you going to find it.
It is not the model learning about you. When a product remembers you properly, your details are not being baked into the model. They are stored separately and supplied alongside your question. This distinction matters, because it is what makes the memory possible to inspect and delete.
Why AI forgets you between chats
By default, nothing gets written down. A language model produces an answer from what is in front of it, and when the conversation closes, that is the end of it. Continuity is not something the model does. It is something a product has to build.
Most AI apps have now built some version of it, which is why you may have noticed one of them recalling a detail from last week. That solves the repetition. It leaves two things unsolved:
- You usually cannot see all of it. The product decided what was worth keeping, and the list is either partial or not shown to you at all.
- It does not travel. What one app knows about you is invisible to every other one. Change apps and you are a stranger again.
Where AI memory is kept, and why that is the important question
There are two arrangements, and the difference is not a technicality.
Memory inside one AI product
The AI company stores what it decided to remember about you, in its own systems, for use in its own product. It works well right up to the moment you want to read all of it, correct something, take it somewhere else, or leave.
A memory layer on your side
The memory is kept separately from any model and handed to whichever model you are talking to, one question at a time. Because it is not the model's property, it can be shown to you in full, edited, exported and deleted, and it works with more than one AI.
That second arrangement is what Tessera is. It is also why we say Tessera is not an AI: it is the memory, not the model.
What good AI memory looks like
- Legible. You can open it and read every fact in it, in plain language, one line at a time.
- Correctable. When something is wrong or out of date, you can fix it yourself in seconds.
- Selective. Only the facts relevant to the question travel with the question. Sending everything, every time, is neither private nor useful.
- Portable. It works across AI models, and you can export it.
- Deletable. Real deletion, of one fact or all of it, without a hidden archive.
- Accountable. A log of every time something read it.
The questions to ask before you let anything remember you
Can I see everything it has kept about me? Can I delete it, for real? Is any of it used to train models? Is there an advertising business that benefits from knowing me? And if I stop using this product tomorrow, can I take the memory with me?
Those five questions separate a memory that serves you from one that serves someone else. Here are our answers to all five.
What it feels like in practice
The change is smaller than it sounds and better than it sounds. You ask for a quick dinner idea and the suggestion already skips meat, because you are off it this week, and leans toward carbohydrates, because you are racing tomorrow. You ask about a gift and the ideas already fit the person, because the AI knows he restores vintage radios and that money is tight this month.
No prompting technique. No preamble. The first answer is the one you would have got on the fourth.
Common questions
Is AI memory the same as a context window?
No. A context window is how much text a model can hold in mind during one conversation. It is working attention, and it empties when the conversation ends. Memory is what survives between conversations. A model with a very large context window still forgets you tomorrow.
Where is AI memory stored?
It depends on the product. Memory built into a single AI app is stored by that company and generally cannot be moved or fully inspected. A memory layer stores it on your side instead, so the same record can be read by you, edited, exported, deleted, and used with more than one AI model.
Is AI memory safe?
That depends entirely on who holds it and what they are allowed to do with it. Ask whether you can see everything it has kept, whether you can delete it for real, whether any of it trains models, whether an advertising business benefits from knowing you, and whether you can take it with you. Our answers.
Do I have to set up AI memory manually?
In Tessera, no. It gets to know you through ordinary conversation, the way a good assistant would in the first week, and turns what matters into facts you can read and change. You can also add or correct anything by hand whenever you want.
Stop explaining yourself to AI.
Ten days free, no card required. iPhone and web.