On being remembered.
Every assistant now wants to remember you. It is the feature that makes them useful: the project you are on, the people you mention, the way you like an answer. It is also the most complete portrait of a person anyone has ever been able to keep, and by default someone else keeps it.
People tell a model what they would not tell a friend. The symptom before the doctor. The debt before the partner. The draft of the letter they will never send. One conversation at a time that was already a lot to hand over. A memory joins the conversations up, and lives on a server you will never see, next to a name and a card.
Forgetting was the only privacy.
Until now the choice has been between two bad options: an assistant that knows nothing about you and makes you start again every time, or one that knows everything and keeps it for its maker, for training, for a reviewer reading transcripts on contract, for the breach, for the subpoena that has not been written yet.
A brain does not work like either. A neuron keeps what it is given in its own connections, and nobody else can read them off. We built Neuron the same way: a memory that is yours in the plain sense, held by your wallet, opened only on your device.
A memory that answers to one wallet.
What you keep is locked on your own device, with a key your wallet makes and nobody else can, before it ever leaves. Once a day the locked pages are bound into a capsule that points at the one before it, copied to three places and checked every day. When you ask something, your device opens the memory, picks the few lines that matter, and sends only those with the question. We hold the pages and never the key. Lose the wallet and they stay shut, for us too.
What it is made of.
- A signature, not an account. Your wallet signs one line of text. We learn an address; we never ask for a name, an email or a card.
- A relay that copies only what it must. Each request is rebuilt from the messages, the model and its settings. Everything else is left behind, and it goes out under one key that every Neuron user shares.
- Routes that keep nothing. Frontier models are reached only through endpoints that retain no prompt, and we say plainly that this part is their word, not a proof. Open models can run sealed in hardware enclaves that publish a signed report.
- Memory you hold. Your conversations and your memories are locked with keys only your wallet can make. Your agents use the same memory, through tokens you can revoke.
- One line of arithmetic. What stays on our side after a request is the bill: which model, how many tokens, what it cost.
What we cannot promise.
A model provider could be lying about what it keeps; that is why every model says how it is reached. When a question goes out with the memories that match it, the model sees those lines while it answers. Your own device is yours to guard, and a wallet linked to your name elsewhere is still linked to your name. Privacy is a set of walls, and we would rather show you where each one stands than tell you the building is safe.
Money that does not ask your name.
Credits and memory are bought with USDC from a wallet, through a program on Solana that anyone can read. It splits each payment in the same transaction: most of it pays for the answers and the keeping, a tenth goes to the people who stake $NEURON, a tenth buys $NEURON on the market and burns it.
Where it ends.
We do not read what you write or what you keep, and we do not train anything on either. Some providers run their own filters and may refuse a request; a refused request costs nothing. What we will not be is a tool for hurting a real person: children, and people’s bodies used without their consent, are outside the door.
Ask anything. Keep what you want kept. It opens for you, and we would like you to check that it is true.
Only you remember.
