Plans
Three levels. Honest prices
We do not sell immortality and we do not promise miracles. We charge for concrete work: storing, serving and supporting what you entrusted to us. Below is what sits behind each number, how one level differs from the next, and what remains yours if you decide to walk away.
“If it is free, ask what pays for it. Put a price on what you sell, and at least everyone knows where they stand.”
CODE: UPRISING, “Nobody's Product”
What you are actually paying for
Memory is built in three layers, and each one costs differently. The price is assembled from them, not invented.
Layer one: the conversation happening right now
Working memory lives in Redis and Vercel KV. This is the context of the current session: what you said three turns ago, and what the assistant concluded from it. The layer is fast and expensive to run, because it keeps data in a machine's memory rather than on a disk.
This is mostly what the monthly fee on Spark pays for. Fifteen dollars is not a membership card for a button. It is the cost of your conversation not starting from a blank page every single time.
The layer works identically on all three levels. The difference between plans begins above it — in volume, in isolation, and in how many people have the right to look inside the archive.
One thing worth saying plainly: the data in this layer is temporary. When a session ends, whatever mattered sinks lower — into the semantic layer and the permanent one. The layer is expensive, but it is not what you end up leaving behind. What actually stays is in the two layers below.
Layer two: meaning, not letters
Semantic memory is pgvector on top of a Neon database. What is stored here is not text but vectors of meaning: compressed representations of what a conversation was about. Because of them the assistant can find the right memory even when you describe it today in completely different words than you used two years ago.
This is also where a common misunderstanding sits. Many people assume that saving memory means keeping chat logs as files and opening them when needed. The files are kept too, of course, but that is only the surface copy. What lets the assistant recall the right thing at the right moment are the vectors in this layer, and vectors have to be computed, stored and kept indexed.
On Family Archive this layer stops being a store and becomes a personal knowledge base. You put your documents, correspondence and notes into it, and the assistant answers from those instead of from general knowledge.
This is where the real line between fifteen dollars and a hundred runs. The first is memory of conversations. The second is memory of everything you decided to put in, plus the right to open it to your family.
Layer three: the part that outlives the company
Permanent memory is Arweave plus a Solana cNFT. Arweave is built so that storage is paid once and up front: the contribution goes into an endowment whose yield pays for storage afterwards. Not a subscription, not a promise — a prepaid volume.
Sixty-five percent of everything passing through the ecosystem router goes to the treasury and is spent buying AR for that endowment. This is not charity and not a gesture. Without topping up the endowment, the word permanent in a description of memory would simply be a lie.
The Solana cNFT here is not decoration and not a collectible. It is a pointer: a record of where in Arweave your archive sits and who it belongs to. The pointer lives on chain and does not depend on whether our servers are running.
Pulling this layer out on its own was deliberate. The question people ask most often is what happens if you disappear one day, and the right answer to that belongs in the structure of the system rather than in a promise. A promise vanishes along with the company; the structure does not.
What you get at every level, including free access
Every user's conversations are saved into a separate folder tied personally to them, on the server and on the blockchain. This works on paid plans and on free access alike, and nobody is charged for it.
The backup runs automatically once an hour, and if a dialog file passes 90 KB it runs immediately without waiting for the hour. No manual clicks anywhere: a mechanism that needs a human to press something will one day fail exactly when it mattered most.
There is no button to press at any step, and that is not distrust of the user but an acknowledgement of reality: people are busy, people forget, people are sure they already pressed it. A mechanism that waits for a click fails on the day it mattered most — so we simply never built the button.
Which gives an honest answer to the question of where the catch is. There is none. Saving memory is how we prove the archive exists and works. The money is charged for volume, for isolation, for family access and for a perimeter of your own — not for the right to leave a trace.
Why the numbers are 15, 100 and 200
Fifteen dollars covers one dialog for one person: the working layer, the semantic layer at a basic volume, and the write into the permanent layer. It is the floor below which storage stops paying for itself and the promise of permanence turns into marketing.
A hundred dollars means several people, several knowledge bases and a markedly larger volume of vectors. A family does not write seven times more than one person, but its archive has to be partitioned: separate access, separate rights and a separate folder for each member inside the shared whole.
A thousand once plus two hundred a month is a separate physical perimeter. The one-time payment closes out the device; the monthly one covers upkeep and a place in the permanent layer. We deliberately did not dissolve the thousand into the subscription: a device is a thing, things are bought once, and paying for one forever is simply wrong.
The relationship between the three prices is not a marketing ladder but a price tag on three different amounts of work. You are free to disagree with the number, but you can see where it came from — and that matters more than the number itself.
“Automatic syncing means exactly that: not one button. If a person has to press something for their memory to be saved, then one day they will not press it.”
— Maksim Galatin, Architect of CODE Eternal
For partners: how payouts are built
Every percentage is named here in full. None of it hides in fine print, because we do not have any fine print.
Where the money from a transaction goes
The ecosystem router on Solana splits every transaction on a fixed schedule: 5 % to the Founder's Fund, 5 % to burn, 15 % to referral level L1, 7 % to L2, 3 % to L3, and 65 % to the treasury, which buys AR for the permanent storage endowment.
The schedule is rigid and cannot be adjusted by hand. The shares add up to exactly one hundred percent, and there is no line item called other expenses.
If there is no referral on one of the levels, that share does not settle with us and is not redistributed among the rest. It goes straight to burn. An empty grid accelerates the token's scarcity instead of feeding an intermediary.
The reason for writing the schedule out in public is simple: the moment the percentages can be changed quietly, any talk of transparency stops meaning anything. Putting them in writing ties our hands, and that is the point.
Ambassador Node: the ordinary user
Any user can register as a node of the network and earn from on-chain transactions of AI assistant memory usage: 15 % on the first level, 7 % on the second, 3 % on the third, paid in $GALATIN.
Payouts are framed as a Network Validation Fee. The wording was not chosen to sound impressive: it describes precisely what the money is credited for and keeps the grid from being mistaken for a pyramid.
Node status has nothing to do with which plan you are on. The on-chain percentages are the same 15 / 7 / 3 on all three levels; the only difference is the base they are counted from, which is covered below in the level alignment rule.
Ambassador Team: a company or a partner with an audience
A partner of this type earns on-chain income on the same terms as an Ambassador Node — the same 15 / 7 / 3 %. On top of that, a two-channel grid opens up on fiat sales of licences and subscriptions: $15 for Spark, $100 for Family Archive, $1,000 once per device and $200 a month for Digital DNA.
Channel one is a fiat payout: 7 % on L1, 3 % on L2, 1 % on L3. Channel two is a payout in $GALATIN through the AIfa Yield Dashboard: 8 % on L1, 4 % on L2, 2 % on L3.
The gap between the channels is not a gift. By choosing the token, a partner triggers an automatic buyback of $GALATIN from the open market for the amount of the reward — that is, they create buy pressure, and the extra percentage is what they are paid for creating it.
So this is not about anyone being favoured: two different actions get two different prices. The rule is the same for every partner, and there are no special percentages to be negotiated.
Level alignment and lost opportunity
Referral income is counted from your own plan, not from the plan of the person you brought in. Sit on the $15 level and a referral on Digital DNA still earns you a percentage of fifteen dollars.
What you missed is not hidden. Your dashboard shows it on a separate line called lost opportunity: exactly how much you did not receive because of the gap between levels. The number is unpleasant, which is precisely why it is kept in plain sight.
Upgrading switches on full payouts from your referrals' subsequent payments at your new level. Nothing is recalculated retroactively — not as a penalty, but as a simple consequence of past payments having already been split along the schedule.
The intent of the rule is straightforward: as your grid grows, you have a reason to raise your own level along with it, rather than earning off the fact that somebody else sits on a higher plan.
The $GALATIN token
The token lives on the Solana blockchain. Emission is hard-capped at ten billion — 10,000,000,000 — and cannot be extended.
Every transaction burns at least five percent, and when referral levels sit empty the undistributed 15, 7 and 3 percent are added to that. The fewer intermediaries in a chain, the faster the supply shrinks.
To be clear: you do not need to hold the token to use any of the three levels. The plans are priced in dollars, and the token is how the referral grid settles up, not a barrier to entry.
We give no investment advice and promise no price growth. What is described here is the mechanics, not a forecast.