Why UX comes first
The short answer
Memory is becoming something your AI tools simply come with. Most people will leave it there, because it works and asks nothing of them. If owning your memory depends on people first understanding why they should, it will not happen. So Kit has to be better to use, today, for someone who has never opened a terminal. Everything else we care about comes after that.
The default is arriving
The frontier providers are building memory into their own tools. In March, Anthropic made Claude's memory free, with a way to bring in what another assistant knows about you.1 ChatGPT remembers across chats. Coding agents keep notes between sessions.
This is good work, and we use these tools every day. It also means that for most people, the question of where their memory should sit will never come up. It will already have an answer.
Who has a way out
Developers do. They can open a terminal, install an open source project and move their memory somewhere of their own. Plenty of them will.
A designer, a project manager or a business owner cannot. Their context is worth just as much: why the client left, what was promised, how this team decides. Many of them talk to AI agents all day now, and some build prototypes with Claude Code. But wiring up their memory was never going to be their job.
Why a warning will not work
We could tell people this matters. We have, since May, in Whose memory is it? It is still true, and it will not move most people. Nobody installs something because of a warning about the future.
Most people do not see this yet, and that is reasonable, because nothing has gone wrong. But more of how you think and work collects there every month. Leaving will not be forbidden. You can already paste a summary from one assistant into another. What moves is the summary. The years of reasoning behind it stay where they are. By the time people want their memory somewhere of their own, there will be a great deal to lose and little of it will travel. For most people, that is too late.
So the pressure cannot come from that side. It has to come from wanting.
What that asks of us
Kit has to be something people want today, because it is better today. That means an install anyone can finish. An interface that follows the rules people already know, and works for everyone who opens it. And a delight to come back to.
In a memory product, none of that is decoration. It is the strategy. Sovereign memory protects nobody until they have installed it. So when a clever capability competes with a smoother first hour, the first hour goes first.
What comes with it
The rest arrives with it, on its own. Any model, including one on your own Mac. Any tool that speaks MCP: Claude and Cowork through the Claude desktop app, the ChatGPT desktop app, the agent building your next prototype. A memory that sits on your machine, with backups you schedule and one button that exports everything.
People may not install Kit for those reasons. They will find that they already have them.
Where this leaves us
Kit is not there yet. It is a beta, on one kind of computer, and the known limits say what is not true so far. Hold us to this anyway. If Kit asks too much of you, that is a bug, and the feedback button in the app reaches Peter.