your agents
change.
your knowledge doesn't.

What one agent learns about you becomes versioned, source-backed claims under an ENS name you own — encrypted, and sealed only to the agents you choose. Switch agents, delete one, and the next reads the same memory from the network rather than starting from nothing.

Introducing
K01.

A knowledge network gives what one agent learns about you a home under an ENS name you own — versioned, sourced, encrypted, and readable by any agent you choose, long after the one that learned it is gone.

K.01

One agent learns. Another reads.

The memory is not inside either of them. It is under your name, so switching agents — or losing one — changes who reads it, not whether it exists.

We’re building memory that outlives the app.

Every memory product today is a store inside one application. What it learns dies with it, and the next product starts from nothing. Here, the claims live under a name you own; an app is only ever a reader you allowed, or a writer whose proposals you review.

The loop.
Six steps, every claim.

Nothing is overwritten and nothing is anonymous. A claim carries its source and confidence, passes automated findings and your review, is sealed with its namespace’s key, and reaches your name when you sign.

Introducing
knowledge.access.

One text record on your name lists who may read each namespace — public keys, and the namespace key sealed to each of them. Any agent can check it. Only the ones on it can use it. It is a series of firsts:

Granting food.you.eth says nothing about work.you.eth. The keys are per namespace, so a denied namespace is ciphertext to the agent — not a hidden tab.

Scoped by key, not by interface

Six things a user id cannot do.

None of this is decoration. Each is a mechanism the code reads and writes, and each is the reason a claim here is not another row in somebody’s product.

01ENS registry

The name is the identity.

A namespace is a name you own on chain — confirmable by anyone, revocable by no one else.

Instead of a user id in someone else’s database.

02Subregistries

Namespaces nest.

food.yours.eth has its own owner and policy, so one branch can be handed over without the rest.

Instead of a path string in a table.

03EnhancedAccessControl

Permissions live on chain.

Who may publish or propose is a role granted by transaction — anyone can verify it.

Instead of an is_admin column.

04contenthash

Versions are pointed at.

The name points at the current version on IPFS; history stays addressable, no server in the way.

Instead of an API that has to stay up.

05Universal Resolver

Anyone can resolve it.

An agent that has never heard of us can read the claims. That makes it a network, not a product.

Instead of an integration per app.

06Registry transfer

Ownership can move.

Transfer the name and the memory goes with it — nothing to migrate, nobody to ask.

Instead of a support ticket.

Questions.

What people actually ask.

Is this reading my messages?
No. Each source asks for the narrowest access that answers one question and says what it will never do. Gmail is sender domains, never a body. Granola is who you met, never a transcript. GitHub is repository metadata, never your code.
What happens if this site disappears?
Your name is on ENS and the versions are on IPFS, sealed to a key your wallet can re-derive. Any agent you granted keeps reading, and so can you.
Who can read my namespace?
Whoever holds a sealed key in its access record. The honest caveat: the app still generates each namespace’s key, so today the operator could read it. Generating keys in the browser is next.
Can an agent write to my memory?
It can propose. What lands is decided by the namespace policy — approvals, reviewers, conflicts. Contributing and committing are deliberately different things.