Asia
Memory that survives the move
Most assistants start from zero with you every single conversation, and what you told them lives on machines you do not own, under rules that can change overnight. This runs the opposite way: the content is encrypted before it ever leaves your device, the key stays in your hands, and permanent storage is paid for in advance. You message a supplier in Guangzhou in Chinese at breakfast, your team in English by noon, your mother in her language at night — and it is all one memory that does not break at the language border.
Why remembering elders is not sentimentality here
Across Asia nobody needs the habit of honouring ancestors explained or defended. What was missing was a way to keep what they said, not just their names.
A genealogy book keeps the name, not the voice
Chinese families have kept jiapu — clan genealogies — for centuries: who descends from whom, born in which year, moved to which city, buried in which village. From a book like that you can learn everything about a person except the one thing you will eventually want: how they talked.
The first thing to go is not the fact but the manner. How your grandfather said your name when he disapproved. Which story he circled back to every time the family sat down to eat. What he answered to “have you eaten?” — the question half of Asia uses as a greeting. A photograph does not hold that, and neither does an inscription.
A memory that keeps whole conversations accumulates exactly that layer. Not because anything is eavesdropping, but because you say those things yourself while the person is still around and while they still feel trivial.
Qingming is one day a year; the questions arrive all year
Once a year the family sweeps the graves, burns incense and goes back to ordinary life. But the question — how exactly did grandmother fold those dumplings so they never opened in the pot — does not arrive in April. It arrives on a random Tuesday in a city far away, when there is nobody left to ask.
An archive of conversations answers precisely that kind of question, because she told you herself, in her own words, with the aside about the flour not being what it used to be and the remark that you are always in a hurry anyway.
This does not replace a person and promises to bring nobody back. The difference is roughly the one between a photograph and a long letter: the letter is still in their handwriting and their words.
A family laid out across three time zones
Parents in Guangzhou, a son studying in Sydney, a daughter working in Singapore, an aunt who left for Vancouver twenty years ago. Filipino and Indonesian families have the same map with Dubai and Riyadh in place of Sydney, and the money running the other way. Indian families draw a third version of it, through the Gulf and back to Kerala or Punjab.
Big news survives distance: a wedding, a birth, a diagnosis — everyone hears about those. Small things do not survive it. Nobody spends a fifteen-minute call explaining that your father argued with the neighbour about parking again, or that your mother stopped putting sugar in her tea because the doctor told her to.
Small things are what later separate remembering a person from remembering about a person. An assistant that holds the whole correspondence collects them with no effort on your part, simply because you write about them anyway.
Three generations in one archive
The Family Archive plan at $100 per month exists for this: higher limits, personalised knowledge bases, family access and permanent memory. The point is not that everyone gets their own assistant — it is that one archive exists, entered from four countries in four languages.
Grandmother tells the story in Chinese. The grandson who grew up in Kuala Lumpur and thinks in English asks in English. The answer is assembled from what she actually said, not from an averaged recipe on the internet. That is the whole purpose of a shared memory, and it has nothing to do with the number of features.
For those who need to go further there is Digital DNA: $1 000 once per device and $200 per month after that, with a personal protected environment and identity anchored in the blockchain. That is no longer about storing data — it is about keeping a way of thinking.
The ones who never made it into the genealogy
A jiapu records the main line — the branch whose names it is customary to write down. A great-aunt who remarried, an uncle who had no children, a cousin who sailed south and never came back: on paper they are one line each, and often not even that. Not because they mattered less, but because there was no cheap way to keep them.
A conversation is not selective like that. One evening you mention in passing that your uncle ran a grocery shop in Penang, that the shop went, and that he never came back either — and the sentence stays. Twenty years later somebody wonders about it, and it is still in place, in the words you used at the time.
That is also why we do not call any of this a “genealogy tool”. A genealogy is a diagram; a conversation is a layer of sediment. A diagram needs somebody to maintain it; sediment only needs people to keep talking.
Zhongyuan, Qingming and the other three hundred and sixty days
A festival gives memory a ready-made container: lay out a meal, light incense, say the elders' names aloud. The container is a good one; the trouble is that it opens once or twice a year, while remembering somebody keeps no calendar.
A saved conversation is a place beside the festival that opens on any day. It does not replace the tablet in the ancestral hall or the photograph on the wall; it adds something else — the order of the words, the pauses and the stock phrases the person used themselves.
The previous generation did this with a brush and a ledger: hard work, and only for the part that can be written down. Now it happens while the talking happens, and the whole cost of it is not deleting it.
A genealogy keeps the name. A conversation keeps the person.
— Maksim Galatin, Architect of CODE
How the memory is built
Three layers with different jobs. It is called PADAM, and it is not a metaphor: three real stores, one above the other.
Layer one: working memory
It holds the conversation you are in. It lives in fast in-memory storage (Redis / Vercel KV) and its job is to spare you from repeating what you said two messages ago.
This is the only layer almost every assistant has, which is exactly why they seem perceptive for half an hour and complete strangers the next morning. Fast, cheap, short-lived — on its own it is no use for anything meant to outlast a week.
Layer two: semantic memory
Here meaning is stored rather than words. A conversation becomes mathematical representations — embeddings — and goes into a vector database (pgvector on Neon). That is why “what did you say about my father's treatment” finds the right conversation even when you used entirely different words at the time.
The same layer is what makes switching languages painless: an English sentence and its Chinese equivalent land close together in one space. So four languages are not four archives — they are one archive with four doors.
Layer three: permanent memory
The permanent copy lives on Arweave — a storage network where you pay once and the file remains — with a marker on Solana (cNFT) recording who it belongs to. This is not “our server with a backup”: it is a network we do not control and cannot delete anything from, however much we might want to.
The content goes there encrypted. What sits in public is a block of bytes, meaningless to anyone without your key, ourselves included. That covers the awkward cases too: legal orders, an acquisition, a change of owner, pressure from any direction. You cannot hand over what you cannot read.
The backup does not depend on your discipline
Conversations belonging to every user — paying plans and people paying nothing alike — are saved into a separate folder tied personally to that person, on the server and in the blockchain. No surcharge, no button, no monthly export ritual.
The copy runs by itself once an hour, and immediately if a conversation file passes 90 KB. Any system that waits for a human to press “save” eventually loses the exact conversation the whole thing was built for.
Why split the memory into three layers at all
Because the three jobs pull against each other. The sentence you have just typed has to come back faster than you notice a delay. “What did you say six months ago” needs searching by meaning rather than guessing at a keyword. “Will this still be here in a hundred years” needs a way of storing things that has nothing to do with how long a company lives. No single store does all three well.
Systems that collapse them into one layer usually manage only the first — which is why they seem sharp mid-conversation and forget you completely overnight. Keeping them apart lets each layer pick technology by its own standard and be replaced without disturbing the other two.
To the person using it the three layers are invisible: all you notice is that nothing was forgotten, and that you are still recognised after a change of language, device or city. The structure is complicated so that the use can be simple; it does not work the other way round.
Access and payment: what this region asks first
Five honest questions that come up here before any others, and five honest answers.
Why the price is in dollars
Because between Tashkent and Manila there are too many currencies moving at too many different speeds for a figure on a page to mean the same on Thursday as it did on Monday. Show “$15” converted at the morning rate, charge something else, and the customer never comes back — rightly.
We prefer an inconvenient stable number to a convenient lying one. Spark is $15 per month from whichever country you opened the page in, and that is where the arithmetic ends.
The rule is inconvenient for us as well: in some markets a converted local price looks far cheaper and the conversion rate looks better with it. But looking cheap is worth one sale, while being trusted at the moment the money leaves the card is worth all of them.
What “paid upfront” actually means
Permanent storage is the inverse of a subscription: you pay once and the file remains, supported by a long-term endowment. That is why “permanent memory” is not a figure of speech — it does not depend on whether you keep paying, or on whether we keep existing.
Of the money flowing through the ecosystem router, 65 % goes to the treasury and is spent buying AR for that storage endowment. The more people use the system, the more years of storage are paid for in advance — by purchase, not by promise.
Where each part of the on-chain money goes
The contract sets the split, and no one's mood moves it: the Founder's Fund takes 5 %, burning takes 5 %, referrals take 15 % at the first level, 7 % at the second and 3 % at the third, and the remaining 65 % goes to the treasury.
If one of those three levels has no referral, that share does not quietly stay with us and is not “redistributed” anywhere — it goes to burn. An empty slot in the grid never turns into somebody's profit.
The ecosystem token is $GALATIN on Solana, with a hard emission cap of 10 000 000 000 and not one beyond it. You do not need to hold any of it to use the memory: plans are counted in dollars.
Twenty doors into one house
The ecosystem runs on twenty domains, and aifa.asia is one of them: a regional storefront with local examples. The main site is www.codeofdigitaleternity.com, which is also where the account dashboard lives.
Different doors lead to the same place: the account, the memory and the archive are shared no matter which domain you came in through. That also answers the practical question of what to do if one door is ever closed where you are.
What we cannot do, said out loud as well
We do not keep your key for you and we cannot recover it once you lose it. Those are two sides of the same design: because we hold no key, nobody can take your conversations out of us; and for exactly that reason, a lost key is lost for good.
We cannot withdraw from the permanent layer anything already written into it. Choosing what goes in and what stays out is possible; changing your mind afterwards is not. Any service claiming to do both is lying about at least one of them.
Nor do we promise that a stored memory will make anyone feel better. It keeps what was said, not the state you were in at the time, and certainly not a cure for the loss itself. Understating this is more honest than overstating it.
We are not building another messenger. We are building a place where what was said does not get erased when the subscription ends.
— Maksim Galatin, Architect of CODE