The memory in your assistant is not your memory
In September 2026, Meta launched Muse and OpenAI announced Dots. Both are agents that remember you, and both are very good at it. If you use either, the experience improves the longer you stay.
That is worth pausing on, because it is also the problem. The memory gets better the longer you stay, and it stays when you leave.
What the memory actually is
Assistant memory is assembled from what you type and what you connect. It is a feature of the assistant, held by the company that runs it, and shaped by what that company finds useful.
To be fair to them, some let you see it. Muse lets you inspect, edit and download its notes about you. That is better than nothing. But a copy of one company's notes about you is not the same as a record you own, and you cannot hand it to a different agent and carry on.
Three things it cannot be
It cannot be built from what actually happened, because it only sees the parts of your life that pass through a chat window. It cannot travel, because it is designed to make that assistant better, not every assistant. And it cannot outlast the company, because it is the company's.
None of that is malicious. It is simply what memory looks like when it is a feature of someone else's product.
The other way round
The alternative is to keep the record yourself, built from your real day, never read by the company that holds it, and let whichever agent you prefer draw on it. Change agent, and the record comes with you.
That is what we are building. Keep the record. Choose the agent.
Also available as source at /blog/the-memory-in-your-assistant.md