Guide
AI assistant memory: what it should remember, and how to control it.
An assistant that remembers that Anna is your sister and that you hate early meetings is far more useful than one that starts from zero every time. It also knows a lot about you. So it is worth knowing what it keeps, and how to make it forget.
In short. Assistant memory is a set of notes the AI keeps about you so it can help next time: people, plans, preferences. It is useful, and it is personal data. Insist on four things: you can see it, correct it, delete it, and turn it off. In the EU, the GDPR already gives you the rights behind those controls, and a proposed EU law would stop AI companions that minors can use from carrying a child’s conversations forward by default.
What memory is for
Without memory, every request starts from zero: you explain again who Anna is, which office you mean, how you like your mornings. With memory, the assistant keeps short notes between conversations, so “move my call with Anna” works without a list of questions. The best memories are small and specific: names and who they are, recurring plans, preferences you stated.
What can go wrong
- It remembers wrong. A misheard name or an old fact stays and colours later answers.
- It remembers too much. A full record of what you said is more than any task needs, and more to lose in a breach.
- It remembers what others told it. An assistant that reads your email also reads strangers; a note planted in a message should not become a memory: prompt injection.
- You cannot see it. A memory you cannot inspect is a profile, not a convenience.
Your rights over it
In the EU, what an assistant stores about you is personal data. Under the GDPR you can ask for a copy of it (Article 15), have it corrected (Article 16) and have it erased (Article 17), and the company must answer within a month: your data rights, with a request to copy.
A proposal goes further for children: the Commission’s KIDS Act would require AI companions and chatbots that minors can use not to carry a child’s earlier conversations into later ones by default. It is a proposal, not law.
Four controls to insist on
- See it: a plain list of what it remembers, in words, not a hidden profile.
- Correct it: fix a wrong fact in one step.
- Delete it: one memory at a time, or all of it.
- Turn it off, for a conversation or for good, and still use the assistant.
And ask two questions: is memory ever used for advertising, and which other companies receive it?
How Lychnos is designed
Lychnos keeps a short memory so it can help: the people you mention, your plans, your preferences. You can read it, correct it and delete it in the app, at any time. The AI receives only the part of your memory a question needs. It is in development: a private AI assistant, by design.
Sources, read 4 November 2026: the General Data Protection Regulation (EU) 2016/679, Articles 12, 15, 16 and 17; European Commission, the KIDS Act explained. General information, not legal advice.
Questions
Can I make an AI assistant forget what it knows about me?
In the EU, yes in most cases: the GDPR gives you the right to see your data, correct it and have it erased, and that includes what an assistant has stored about you. A good assistant also lets you delete memories yourself, one by one.
Should an AI assistant remember my conversations?
Only what helps you, and only where you can see and change it. A short memory of people, plans and preferences is useful; a full record of everything you said is rarely needed.