Portable memory
One private memory.
Any compatible AI.
Most AI memory belongs to the assistant, not to you. Switch models and it's gone; switch apps and you start over. Portable memory inverts that: the memory is yours, it lives with the intelligence layer, and interfaces borrow from it.
The trap
The memory is the switching cost.
After a few months of use, an assistant knows your preferences, your projects and your people. That accumulated context — not the model quality — is what makes leaving expensive. And it is exactly what you cannot take with you.
Keeping memory in a layer above the models means changing model is a routing decision, not a migration.
What travels with you
Preferences
How you want answers: length, tone, formatting, how much hedging you'll tolerate.
People
Who matters in your work, and what their relationship to you is — tokenised before it travels.
Projects
What each project is, where it stands, and the decisions already made.
Standing instructions
Rules you shouldn't have to repeat: units, currencies, house style, things never to suggest.
Decisions
What was chosen and why, so future answers stay consistent with it.
Writing style
Samples and patterns that make drafts sound like you rather than a model.
Three storage choices
Cloud, this device, or off.
Cloud memory travels between your devices and can reach a connected application. Device memory is encrypted in the browser and never uploaded unless you explicitly ask. Off means Kelven keeps nothing persistent about you at all — conversations still work. Records marked local-only stay on the device even in cloud mode.
Kelven app
Complete experience
Claude
In testing
ChatGPT
Experimental
Other MCP clients
Compatible
The intelligence layer
Kelven
Routing, memory and privacy rules live here — not inside any one interface.
Kelven controls what Kelven shares. It can’t hide something you typed directly into another app.
Portable, not exposed
Borrowed context, minimised each time.
A compatible MCP client never receives your memory store. It asks a narrow question, and gets back the few facts the task needs — filtered by sensitivity, trimmed, and tokenised so identities become typed placeholders instead of real names.
You can see every request, what was returned and what was withheld, and cut an app off in one click.
The future isn’t another AI model. It’s an intelligence layer that’s yours.
Kelven is the user-controlled intelligence layer between you and AI — with one memory that’s always yours.
No spam. Just one note when early access opens.