There is a war on for your context, and you are not invited to the briefing. Every major AI lab is now shipping some version of the same feature: the assistant remembers you. ChatGPT has memory. Google's AI products remember across surfaces. Anthropic, xAI, and the rest are racing the same direction. The framing is always convenience. The underlying bet is identical: whoever owns the running record of what you think about owns the relationship.
This is not a feature war. It is a land grab. And the land is you.
The players, in one paragraph
OpenAI moved first with persistent memory in ChatGPT and is now folding it into agent-style workflows where the assistant keeps state across tasks. Google has the deepest moat — Search, Gmail, Docs, Calendar, Android — and is stitching all of it into a single personal context layer that no competitor can match on raw surface area. Anthropic is more cautious on memory but is building toward long-running agents that effectively accumulate context by doing. Apple is the dark horse: late, but with on-device privacy as a wedge and a hardware distribution channel nobody else has. Meta wants memory too, but mostly so it can sell better ads against a richer model of you. Everyone is converging on the same shape: a persistent, personal context store that makes the assistant hard to leave.
What "winning" actually means
Memory is the stickiest feature ever invented in software, and the labs know it. Once an assistant knows your projects, your taste, your recurring questions, your kids' names, and the way you phrase things, switching costs stop being measured in dollars and start being measured in months of re-explaining yourself. You don't churn from a tool that has a year of you inside it. You just don't.
That's why every lab is willing to give memory away for free. The feature is not the product. The retention is the product. The data is the product. The optionality on what to do with both of those later is the product. You are not the customer. You are the substrate.
The three bets they're all making
Strip away the branding and the entire industry is making the same three bets at once.
First, that you won't read the policy closely enough to notice that "product improvement" is an unbounded category. Second, that you won't care enough to leave once the lock-in compounds. Third, that regulators will move slowly enough that the asset — your context, accumulated at your expense — gets big enough to defend before anyone builds a fence around it.
These are not crazy bets. The first two have worked for every advertising-supported platform since 2008. The third has worked for every one of them at least once. The labs are not stupid. They are doing the rational thing for companies with their cost structures. Your interests and their interests just happen to diverge at exactly the point where the data gets valuable.
Why "free memory" is the most expensive thing on the menu
If a company is giving you a feature that costs real money to run — storage, embeddings, retrieval, inference on every turn — and not charging you for it, you are not the customer. The feature is being subsidized by something, and in this industry that something is always downstream use of the data. There is no other unit economics that survives.
This is the part people miss when they compare memory features on a checklist. "ChatGPT remembers for free" and "Eigen Mesh costs €3.99 a month" are not the same kind of comparison. One is a product you pay for. The other is a product where you are the inputs. The first has a price. The second has a cost you can't see and can't audit.
The thing nobody is building
Here's what's strange about the whole race: almost no one with serious funding is building memory that you own. The labs are building memory they own and let you use. The open-source tinkerer kits — mem0, OpenBrain, the various RAG-on-a-laptop projects — are building infrastructure for people who want to assemble their own stack from parts. There is almost nothing in the middle: a real product, that you pay for, that you can also run yourself, that does not treat your context as raw material.
That gap is why Eigen Mesh exists. It's an open-source (Apache 2.0) personal AI memory — a second brain you own, portable across whatever AI tools you use this year and next year. You capture what you think and plan, deliberately, and Eigen Mesh organizes it and offers it to whatever assistant you happen to be talking to. The assistants come to your memory. Your memory does not get uploaded to theirs.
It is not a framework. It is a product. You install it, or you use the hosted version, and it works. You don't assemble anything. And because it's Apache 2.0 and self-hostable, the question of whether you trust us is one you can answer by running it yourself.
Why you should care, even if you don't care
You can ignore this war and probably be fine, the way you can ignore most platform wars and probably be fine. The cost shows up later. It shows up the day the lab changes its pricing, or its policy, or its model, and you realize the year of context you accumulated is not yours to take with you. It shows up the day a feature you depend on gets gated behind a tier you don't want to pay for. It shows up the day "product improvement" quietly expands to include something you would not have agreed to if it had been spelled out.
The fix is not a better policy from a better company. Companies change. Policies change. The fix is a different place for the memory to live. Your memory, not theirs.
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 to start, then €3.99/month — less than a coffee. Servers in the EU. We don't train on your data.