Picture this

Who it is for

Tunde moves from one AI assistant to another and loses nothing: his preferences and notes sit in an encrypted vault he controls, and he grants each app a revocable key.

Two ways to own it

What you get

Base$299

$75 deposit to reserve

Encrypted vault, grant and revoke registry, MCP connector, access log

You choose: chain
You choose: name and logo
You choose: accent color

Customfrom $2,000

Built around you

Reshape it entirely: new features, your flows, your brand end to end, integrations you need. We scope it with you.

Everything in Base
Custom features and design
Priority build

Why it is new

What exists today, and what this does differently

Why own your own

User-owned AI memory with revocable onchain grants has live and hackathon implementations.

Closest things out there

SealedMind claims a live 0G mainnet memory vault with onchain revocable grants, alongside Mnemo, EchoVault and OpenGradient's user-owned memory vision.

Checked with web searches in October 2026. A search is not proof that nothing similar exists, and the market moves fast, so we recheck before every build.

What it unlocks

The human in it

Your preferences and history become yours, not the platform's.

The Regatta Board

Races it can enter

grant · NEAR Foundation · rolling

NEAR Foundation / NEAR AI grants

unknown

Official page ↗

User-owned AI theme; eligibility unknown

Their rule on existing code: unknown

grant · Ethereum Foundation · rolling

Ethereum Foundation ESP

rolling

Official page ↗

Public goods angle; existing projects OK

Their rule on existing code: Existing projects OK

What you own

The whole app, not a starter kit.

The complete Keepsake Mind app: people keep what AI assistants learn about them in an encrypted vault they own and lend each app a scoped, revocable key, with a paywall, owner console, MCP connector and importer built in.

Base price · $299

The user app, complete

Every buyer gets the same finished app. Basic branding and copy changes come through a quick form.

Three step onboardingHome

Shown once: what AI learns is yours, shelves not one pile, keys you can take back.

Progress and next actionHome

Imported, new key, old app revoked and tidied, with one clear next step.

Shelved encrypted vaultVault

Five shelves with counts, sensitive markers and key duration choices.

Distilling chat importVault

Imports years of chats and keeps only stable facts. Gated behind Plus, with a free starter list.

Per read approvalsVault

Family and Money ask for a tap on every read. Gated behind Plus.

Grant, scope, deny and revokeKeys

Three apps knock; grant a timed key, scope one to a single shelf, deny or revoke, with a key receipt.

Free key limitKeys

The free plan holds a set number of live keys; more unlocks with Plus.

Access logLog

Every read with app, shelf and time, refused reads, and a daily root on Base.

One tap sensitive readsLog

Allow or refuse a single Family read from an app with a key.

Sunday tidyTidy

Apps only propose memories; keep or forget each one, with a streak and share card.

Plans and settingsMe

Pricing from the owner paywall with crypto checkout, passkey, export and share card.

Open the full app demo

Add-on

Owner console

Your admin side: revenue, paywall editor, users, rules and the agent log.

Revenue and paywall editorRevenue

MRR from Plus and app makers, payouts, incoming USDC, payment methods, and live editing of tiers, prices, free keys and gates.

App maker listingsMakers

List assistants on the MCP connector, decline over asking ones, per active user billing.

Rules, agent log and roadmapRules

Max key length, import threshold, sensitive defaults, log anchoring, pending writes; plus the Agent log and priced Roadmap.

Top-up

Growth and referrals

Referral and growth mechanics, switched on when you are ready to grow.

Switch weeks and referralsGrowth

Run a switch week with a partner assistant, see loops and simulate referrals.

The money path, built in

How it earns for you

Owning your memory is free. People who want more keys, per read approvals and full imports pay for Plus, and assistant builders pay per active user to offer instant personalisation through the connector.

PersonalFreeOne vault, up to 3 app keys, the access log and the Sunday tidy

Plus$4 a monthUnlimited keys, per read approvals on sensitive shelves, full chat import from any assistant

For app makers$0.20 per active user a monthAssistant builders offer instant personalisation from a user's vault via the MCP connector

Base gas for a grant or revoke is usually under one cent; the owner can sponsor it.

Prices are yours to set. Illustrative, not a promise of income.

Growth loops, built in

How it spreads

Import from your vault buttons inside every connected assistant
The it still knew me share card after a switch
Builders adopt the connector to cut cold start
Switch weeks run with partner assistants
Friend referrals worth a free Plus month for both

Roadmap · build the next phase with us

Where it goes next

v1 included

The full memory vault

Encrypted shelves with passkey login
Grant and revoke registry on Base
MCP connector for any assistant
Distilling importer and Sunday tidy
Access log, paywall, owner console

Included

Builder SDK and NEAR

Drop in SDK for assistant builders
Usage metered billing per active user
NEAR and Arbitrum registries
Family vaults with shared shelves

Add-on from $600

Portable agent identity

Memory for autonomous agents
Selective disclosure proofs
Enterprise vaults for teams
Inheritance and recovery contacts

Add-on from $1,400

For investors and grant reviewers

The pitch, ready

Your AI memory in a vault you own, lent to any assistant with a key you can take back.

Problem: People spend years teaching AI assistants how they write, what they eat and who their clients are. That memory lives inside each app, so switching means starting from zero and the app decides what happens to it.

Solution: Memories are encrypted on device into shelves. A small registry on Base records which app holds a key to which shelves until when; an MCP connector lets any assistant read granted shelves and only propose new memories. Every read is logged.

Why now: MCP gives assistants a common way to call tools, passkeys remove seed phrases, and people are switching assistants more often than ever.

Market: Heavy AI assistant users, and the assistant builders who want day one personalisation without hoarding data.

The ask: Two assistant builders as launch partners and grant support from Base and NEAR ecosystem funds.

Funding is never guaranteed. Check each program's rules.

For hackathons and demo days

Your 2 minute demo

Home: meet Tunde in Toronto at 1 AM, leaving Parrot AI after three years.
Vault: import 1,940 chats, keep 221 stable facts on five shelves, or hit the Plus paywall.
Keys: give Lumen a 30 day key, scope FreeChat to one shelf, revoke Parrot.
Log: every read with app and shelf, then allow one Family read with a tap.
Tidy: keep or forget three proposed memories and share the switch.
Owner Agent: the importer's two failures and how they were fixed.
Owner Revenue and Makers: change the Plus price and maker rate, list a cooking assistant.
Owner Growth and Roadmap: run a switch week, then request the builder SDK.

Many hackathons require code written at the event. See the Regatta Boardfor each race's rule.

Make it entirely yours

Custom work from $2,000: your own assistant integration, enterprise policies, a different chain or key service, or a branded deployment. Scoped on a call.

Custom from $2,000

BLUEPRINT

Why it matters and how it really works

The need

People now spend years teaching AI assistants how they write, what they eat, who their clients are. That memory lives inside each app, so switching means starting from zero, and the app, not the person, decides what happens to it.

What exists and what is new

This already exists in several forms. SealedMind claims a live memory vault on 0G mainnet with onchain revocable grants, and Mnemo, EchoVault and OpenGradient's user-owned memory work explore the same idea. Research verdict: exists, low novelty. What this build adds is a packaged, owner-ready version: shelf-level keys (so a family fact cannot leak into work), timed grants, an access log a person can read, a distilling importer, and a weekly tidy habit, on Base with an MCP connector any assistant can use.

The data it runs on

Memories are short facts, each with a source chat, date and shelf, encrypted on the user's device before upload. Onchain: the vault root hash, and per grant the app, shelves, expiry and revocation. Read events are logged offchain and signed, with a daily hash anchored onchain.

1.The mission

What an AI learns about you belongs to you, and goes where you go.

2.Why people can't resist it

Day one, it knows you A new assistant reads your vault on first launch, so switching costs nothing.

Keys you can take back Every app holds a scoped, timed key. Revoke in one tap and its next read fails.

Sunday tidy Ten minutes a week to keep or forget the new memories apps wrote. A streak keeps it light.

3.How it works for real

Wallet and identity: Privy embedded wallet on Base. Passkey login, no seed phrase; the wallet signs grants and revokes.

Encrypted vault: Memories encrypted client side with per-shelf keys (for example Lit Protocol access conditions or a simple envelope scheme), stored on IPFS or a private bucket. Only ciphertext leaves the device.

Grant registry: A small contract on Base maps app address to shelves and expiry. Apps prove a valid grant to the key service to get a shelf key; revoke flips the entry.

MCP connector: An MCP server any assistant can call: read memories for granted shelves, propose new memories (which land as pending for the owner to keep or forget).

Distilling importer: Imports exported chats and keeps only stable facts using a cheap model tier (for example Claude Haiku or a comparable small model via OpenRouter), with a mid tier for ambiguous ones.

Access log: Every read is logged with app, shelf and time, signed by the key service, with a daily Merkle root written to Base.

4.Cost per use and pricing

Import: about 2,000 chats at a few hundred tokens each through a small model tier is roughly $0.50 to $2 one time. Reads: no model cost, only key service and storage, well under $0.001 each. Base gas for a grant or revoke is usually under one cent. Example month for 1,000 users: storage and key service about $40, daily log anchoring about $3, imports for new users about $100, hosting $30, so about $175.

Personal Free

One vault, up to 3 app keys, the access log and the Sunday tidy.

Plus $4 a month

Unlimited keys, per-read approvals on sensitive shelves, full chat import from any assistant.

For app makers $0.20 per active user a month

Assistant builders pay to offer instant personalisation from a user's vault via the MCP connector.

5.Growth loop

Every app that adds the connector advertises import from your vault, which pulls users in from rival assistants. Users share the switched assistants and it still knew me card. Builders adopt it because it cuts cold start.

6.Risks and guardrails

Copies already made: Revoking stops future reads, not copies an app already kept. Say so plainly, prefer apps that sign a no-retention pledge, and show it in the grant screen.

Sensitive data: Family, health and money facts are high risk. Default those shelves to ask every read, keep them out of import unless starred, and follow GDPR, PIPEDA and NDPR style rights to delete.

Key service as a weak point: If the key service is breached, shelf keys could leak. Use threshold or hardware-backed key management and rotate keys on revoke.

Bad memories: Distilling can save wrong facts. Every memory shows its source chat and can be edited or forgotten; apps can only propose, not write.

Existing competition: Several projects do similar things. Compete on usability, the connector and the weekly habit, not on novelty claims.

7.Build order

Client side encrypted vault with five default shelves
Grant and revoke registry contract on Base
Key service that checks grants and releases shelf keys
MCP connector with read and propose memory tools
Chat importer that distils stable facts
Access log screen and Sunday tidy

8.Where this is going

Now
One person moves assistants and keeps 221 memories, with keys he can take back.
Next
Assistants compete to be granted a key, and advertise what they will not retain.
Later
Your vault follows you into every agent, shop and service, each seeing only the shelf it needs.
The world it builds
Personal context becomes something people own and lend, not something platforms hoard to lock them in.

Costs and tools are estimates at the time of writing. General information only, not legal or financial advice.

OWNER PLAYBOOK

Own it, then chart your voyage

Nine routes from one vessel. Tap a port to open its chart.

Why own Keepsake Mind

Every AI assistant you use is quietly building a memory of you that it keeps when you leave. When you own Keepsake Mind, your preferences, notes and history sit in an encrypted vault under your wallet, and each assistant gets a key you can revoke. Run it under your brand for a community and you become the trusted place people keep their AI memory, not another platform harvesting it.

What’s stopping you?

“I can't code. Can I run a memory vault?”

Yes. The vault, the grant and revoke registry and the MCP connector arrive deployed. You add notes, connect an assistant that supports MCP, and grant or revoke from a screen.

“Is my data really private?”

Memory is encrypted before it is stored and only keys you grant can read it. Onchain you store grants and the access log, not your notes. No system is perfect, so do not store passwords or seed phrases, ever.

“SealedMind and others already exist. Why own one?”

Honestly, user owned memory has live projects. Owning yours means your brand, your chain, and your community's trust, and you are not dependent on one vendor staying alive. If you just want a personal vault, a hosted one may do.

“What if an assistant does not support MCP?”

Many assistants now do, and the list grows. For ones that do not, you can export a context pack and paste it. Be honest with users about which assistants connect natively.

“Is storing people's memories a legal risk?”

If you host others, you are handling personal data, and laws like GDPR may apply. Keep data encrypted, let users delete it, write a plain privacy page, and get local advice before charging at scale.

The million-dollar ladder

Individuals pay a monthly plan for vault storage, connectors and access logs; teams and communities pay for shared vaults where members keep private memory and grant a shared assistant limited access. This is a target, not a promise.

$50

$4,000

$30,000

$85,000

First 10 users

10 heavy AI users you know

$5 per month

Power users

500 users

$8 per month

Team vaults

3,000 users

$10 per month blended

Million-dollar run rate

8,500 users

$10 per month blended

Connector count: every new assistant supported is a new reason to join
Team plans for small companies worried about staff pasting context into random tools
An access log users can show, which builds trust and word of mouth
Annual plans for people who switch assistants often

Illustrative stages, not a forecast. Results depend on you and the market; nothing here is guaranteed.