KEPTmind
01What won't change
  • Models change.
  • Assistants change.
  • Companies change.
  • What you know doesn't.

Your knowledge is the one asset that outlives every product built on top of it.

The next assistant will be smarter than this one, and the one after that smarter still. What makes any of them useful to you is the same thing every time: what you know, what you've decided, what you've been through. That should never be the thing you lose when you switch.

The memory your AI keeps about you should be kept by you.

Every AI assistant is asking to read your life. Keptmind is the place they read from — encrypted, every read logged, revocable in one switch. And what they learn about you is written back to you, in plain text you keep forever.

A proposal · 2026One person's knowledge, set in typeVault open

One person's knowledge: family, home, health, recipes, ideas, work, money, the unfinished novel, and everything they have told an AI assistant.

02What every assistant asks for

Muse · Grok Bot · Instinct · OpenClaw · ChatGPT · Gemini · Claude · Copilot · Alexa+

Every one of them asks for the same thing:

Access

your emailyour calendaryour filesyour notesyour messagesyour healthyour moneyyour photosyour history

Then it remembers —

in its system, in its format, under its terms. The better it knows you, the harder it is to leave. That memory is the moat.

Leave, and your memory stays behind.

You hit a wall. Starting over with a new assistant means teaching it your life from zero — or handing everything over again, to someone new.

03How it works — try it

One instrument between your life and every agent.

Pick an agent. Flip what it may read. Watch every read happen. Sealed rooms ask you first. What the agent learns is written back to your vault — and one button ends it all.

KM—1Personal knowledge controller
Reading

Agent01

Ends this agent's access

May read02

Written back to you04

03 · Every read0 reads

    A simulation. Nothing here leaves your browser.Try: switch on health/, or press revoke.
    04Eight promises
    1. 01

      Plain text, forever

      Markdown and simple files. Readable in fifty years, by any tool, with or without us.

    2. 02

      Written back to you

      Whatever an agent learns or produces about you lands in your vault — not only in theirs.

    3. 03

      Least access

      Trying a new agent? Give it three folders, not your life. Scope by folder, file, time and purpose.

    4. 04

      Every read, seen

      Who read what, when, and why — a complete ledger, in plain text too.

    5. 05

      Sealed rooms

      Medical records, money, legal. Nothing enters without your yes, case by case.

    6. 06

      A watchman

      An agent sweeping everything it can reach gets paused and reported to you.

    7. 07

      One switch

      Cancel an assistant and its access to your vault ends with it. No wall, no hostage.

    8. 08

      Not our business

      We sell protection, never access. No ads, no training on your data, no data brokering.

    05Who holds the key

    You choose who holds the key. We show you the price of each choice.

    I

    Only me

    You hold the only key. We store ciphertext we cannot read. Agents work only while your device unlocks them.

    Privacy ●●●Convenience ●○○Lose the key, lose the vault
    II

    Me + a timed key

    You issue a derived key that expires — an hour, a week — so agents can work while you sleep. It dies on schedule.

    Privacy ●●○Convenience ●●○Bounded by time and scope
    III

    Me + recovery

    Recovery through people you trust, or held in escrow, so a lost device isn't a lost life.

    Privacy ●○○Convenience ●●●Recoverable

    This is where we will spend our research. The line we won't cross: our business is protecting your data — never being the one who hands it out.

    06Who it's for, first
    You build with agents

    One vault, every agent

    You run Claude, ChatGPT, local models and your own scripts over the same notes. Point them all at one vault — over MCP or as a folder — with scopes, a ledger, and memory that doesn't fork per tool.

    mcp://vault/work/** · read
    memory/claude/decisions.md · write
    You carry your own records

    Your health, your file

    Three clinics, two countries, one history. Keep your records in a sealed room, and let a health assistant read one visit — once — with your yes.

    health/2026-03-visit.md · sealed
    asked by health-coach · allowed once
    You're trying something new

    Try it, then trust it

    A new assistant wants everything. Give it three folders for a week, read its ledger, and decide. Walk away, and what it learned stays with you.

    trial: travel-agent · 7 days · 3 folders
    expires 2026-10-05 · auto-revoke
    07The honest version

    What we promise

    • Your knowledge lives in open, plain formats you can take anywhere.
    • Every read by every agent is recorded, and the record is yours.
    • Access is scoped, time-bound, and ends the moment you say so.
    • Sealed rooms never open without your case-by-case approval.
    • You never hit a wall: leave any assistant and keep what it learned.

    What we can't

    • Un-copy what an agent already read. We limit, log and cut — we can't reach into another company's servers.
    • Make "only you hold the key" risk-free. If you choose it, losing the key means losing the vault.
    • Force every provider to connect. We start with open agents and open protocols.
    • Ship tomorrow. This is a proposal — we are asking before we build.
    08The film
    09Questions
    What is Keptmind?

    A vault for your personal knowledge that AI agents read from and write to — with your permission. Everything is stored as plain text you own, every read is logged, and any agent's access can be revoked in one switch.

    Can Keptmind read my data?

    You choose. You can hold the only key, or allow a time-limited delegated key so agents can work while you're offline. Either way, our business is protecting your data — never brokering access to it.

    What happens when I stop using an assistant?

    Its access to your vault ends the moment you revoke it or cancel. Everything it learned about you stays in your vault, in plain text, ready for the next one.

    Can you stop an agent that already copied my data?

    No, and we'll never claim to. We can limit what it reads, show you every read, flag bulk reads, and cut future access. What a provider already copied is governed by their terms and the law.

    How do agents connect?

    Through open protocols like the Model Context Protocol (MCP) and a plain filesystem view, so any agent that can read files or speak MCP can use your vault — with scoped permission.

    Is it available?

    Not yet. We're asking first. Join the list to shape what gets built, and to be invited first.

    10Nothing is built yet

    Keep your mind.

    We're asking before we build. Tell us who you are and what you'd keep first — every answer shapes what gets built, and the first people on the list are the first invited.

    We store your email and your answers. Nothing else, no tracking cookies. Privacy.

    Which assistants do you use — or are trying?

    Personal agents

    Chat & work assistants

    Your own

    What would you keep first?
    Would you pay for it?

    One email when there's something to try. Unsubscribe anytime.

    You're on the list
    №

    Thank you. Your mind, kept.

    If this matters to you, the best thing you can do is send it to one person who would care.