The same person is an employee, a consultant to two clients, and someone with their own project — often in the same afternoon, across four different tools. Every one of those boundaries is a place context gets lost, or worse, leaks.
This page is the argument you can touch. Switch a hat and watch what seals. Switch a lens and watch what recedes. The one thing that never happens: nothing moves. The substrate is constant — only the projection changes.
A working week's context, laid out once. Positions are fixed and never animate — if a hat switch moved things, it would be a filter, not a boundary. Try All hats first, then any single one.
They are different needs and they get different treatment. Conflating them is how a context platform becomes a leak.
Employer, client A, client B, personal. Switching hats removes — objects in other hats are not dimmed, they are not there. An agent working in this context cannot see them, cannot search them, and cannot be persuaded into them.
Reviewing, implementing, deciding. Switching lenses re-weights — everything stays present and connected, but what matters right now comes forward. No object is copied, moved, or hidden.
The claim is one word away from a compliance failure, so it has to be exact about what is tied.
The tie is at the infrastructure layer. Never at the content layer.
— One substrate, many bounded contexts — not "one place for everything"There isn't one. And this is not a gap a memory feature can close.
Platform-native memory models "the assistant remembers you." One continuous self, owned by the platform, learning across everything you do in it.
The better it learns, the more it carries between your roles — and it has no primitive for saying which you it is talking to, nor any way to prove separation to somebody who asks.
Roles are the primitive, not an afterthought. Separation is enforced at the substrate, so it can be demonstrated rather than promised.
Closing this gap on their side would mean rebuilding as user-owned, bounded, and provably unread. That is not a feature they can ship; it is a different product.
Yes — and that is exactly why the zero-knowledge posture is load-bearing rather than a nice-to-have. Concentration is only safe if the thing concentrating cannot read what it holds.
A switchboard, not a reader. It connects your contexts without ever being able to look inside them.
— Bodies stay encrypted; there is no server-side body index, and there is not going to be one