You tell ChatGPT about your project. Your deadlines. The client you can't stand. The side hustle you're nervous about. You tell it things you wouldn't put in an email. And it remembers — for you, supposedly. The catch is the word "supposedly."
The memory feature in tools like ChatGPT and Google's AI products is sold as convenience. It is also, quietly, one of the most valuable datasets ever assembled: a running transcript of what you think about, worry about, and plan to do. That data does not live in your notebook. It lives on their servers, under their terms, shaped by their roadmap.
Memory that works for them, not you
When ChatGPT "remembers" that you're a freelance designer with a toddler, it's not doing you a favor. It's building a profile. That profile makes the product more useful to you — and more useful to the company, to advertisers, to whoever they sell the next tier of access to. The memory feature is a two-way mirror. You see a helpful assistant. They see a high-resolution model of your interior life.
This isn't paranoia. It's the business model. The big AI labs are not charities. They are capitalized at staggering valuations on the premise that they will turn attention and context into revenue. Your context is the product. The "memory" button is a consent flow dressed up as a feature.
The asymmetry nobody talks about
Here's the part that should bother you more than it does: you can't take it with you.
If you spend a year training ChatGPT to know your taste, your projects, your recurring questions, and then you switch tools — or get banned, or the company changes its policy, or the free tier gets worse — all of that is gone. You don't export it. You don't own it. You were renting intimacy from a company that can evict you at any time.
Worse, you trained it on yourself and got no equity in the result. The model got better at being a personal assistant. You got dependency.
What "your memory, not theirs" actually means
We built Eigen Mesh around one stubborn idea: the things you think and plan should belong to you. Not to OpenAI. Not to Google. Not to us, really — you can run it yourself.
Eigen Mesh is an AI-native second brain. You say what you're thinking or planning — a meeting outcome, a half-formed idea, a goal for the quarter — and it remembers, organizes, and makes that context available to whatever AI tool you're using. Your memory sits in one place that you control, and your assistants come to it. Not the other way around.
It's the difference between a notebook and a surveillance system that's polite to you.
The technical difference, in one sentence
Most AI memory mines your emails, transcripts, and PDFs to infer what matters. Eigen Mesh works the other way: you say it, you meant it, it's saved. No inference about what you "probably" care about. No quiet indexing of your inbox. Just the things you chose to keep.
That's a different kind of dataset. Smaller, sharper, and yours.
The thing about "anonymized"
The standard defense is that the data is anonymized, aggregated, used only in the aggregate. This defense has been worn thin by every industry that's ever made it. Aggregates leak. "Anonymous" profiles get re-identified. And the more granular the underlying data — and a memory transcript is about as granular as it gets — the easier re-identification becomes.
Even if you trust the company today, you're trusting every future version of that company, every future acquisition, every future change in leadership and policy. A dataset that lives forever on someone else's servers is a dataset that will outlast whatever promise was made when you agreed to share it.
Why this matters now
Memory features are being shipped as fast as they can be built because they're sticky. Once an assistant knows you, you don't want to start over. That's the point. The lock-in is the feature, not a side effect.
The question isn't whether AI should remember things about you. It should. The question is who holds the memory, and on whose terms. Right now the answer is "a company you don't control," and that answer is going to age badly.
You have two options. Keep renting your context from someone who can change the lease whenever they like. Or keep it yourself, in a system you can run, fork, or walk away from.
We picked the second one. If you did too, come build with us.
Eigen Mesh is open source under Apache 2.0 — github.com/hikaru90/eigen. Self-host with Docker Compose, or get early access to the hosted version. 100 free credits, then less than a coffee a month. Servers in the EU. We don't train on your data.