Permanent memory is not a button. For it to work as intended you need to understand three things: where each thing is stored, who holds the key, and what actually happens when you delete. This is that, taken apart step by step — from the first conversation to reading your own referral grid.
From how memory is built to your own protected perimeter.
Working memory holds the current conversation. Semantic memory holds experience sorted by meaning. Permanent memory is immutable storage. Why each exists, what breaks without it, and why you cannot keep only the third.
Data is encrypted before it reaches permanent storage. Deletion destroys the key rather than writing “deleted” in a database. What that means in practice, what it costs, and why there is no way back.
Your conversation folder, your plan and the next charge, the invitation link, the three-level grid and the lost-opportunity line. What each figure means and what it changes.
Fifteen, seven and three per cent across three on-chain levels; for company partners, an extra fiat channel and a $GALATIN buyback channel. The level-alignment rule, in full.
What Digital DNA includes, how a personal perimeter differs from a shared one, and when it is genuinely required rather than money spent on something you will never use.
Forty-two checks: what each one means, which findings are serious and which are cosmetic, and where to start when you cannot fix everything at once.
PADAM is not a metaphor. It is three separate stores with different speeds, costs and lifespans.
AI memory usually fails in one of two places. Either it is too short — the assistant remembers the last few messages and forgets everything from yesterday. Or it is too expensive — re-reading the whole archive on every request runs straight into the size of the context window and into the compute bill.
Splitting memory into three layers solves both at once. The fast store keeps what is needed right now and costs almost nothing. The semantic store finds the right piece of the past without re-reading all of it. The permanent store makes sure the past still exists in twenty years — and takes no responsibility for speed or search.
Most comparisons make the same mistake: they line the layers up and ask which is best. They do not replace each other. Remove the first and the conversation falls apart within a single session. Remove the second and the assistant stops recognising you. Remove the third and everything you have accumulated lives exactly as long as the company holding the servers.
Technically this is a key-value store held in RAM: Redis, in our case through Vercel KV. It holds the context of the conversation you are having now — the last exchanges, who you are, what is being discussed, which clarifications have already been made so you do not have to repeat them.
The defining property of this layer is speed. A lookup takes milliseconds, which is why the assistant answers without pauses. The second property is impermanence. RAM is not built for decades, and treating it as an archive is the most common beginner's mistake.
The practical conclusion: if a conversation ran for hours and mattered, do not rely on the first layer to hold it. It is not supposed to. Two other layers exist for exactly that, and the handover happens without you doing anything.
This is a database with vector search: pgvector on top of Postgres in Neon. Every meaningful fragment of experience becomes a vector — a set of numbers describing what the fragment means rather than which letters it contains. Things close in meaning end up close together in that space, even when written in completely different words.
This is where the effect everything was built for comes from. You ask “what did we decide about that whole business with the move”, naming no dates and no names. Keyword search finds nothing — there is nothing to match. Meaning-based search finds it, because it searches by proximity of sense.
The second layer also handles the economics. What goes into the model is not the whole archive but the few fragments closest in meaning. That is why the size of accumulated memory has almost no effect on the cost of an answer: the storage grows, the per-conversation bill does not.
This layer has an honest limitation worth knowing in advance. It searches by meaning, and meaning is approximate. Ask something too broad and too much matches, and the answer comes back vague. A precise question almost always gets a precise answer. It is a skill, and it takes about a week to pick up.
Arweave is a permanent storage network where you pay once and storage is prepaid for a horizon measured in generations rather than years. A Solana cNFT — a compressed non-fungible token — acts as the receipt: it ties a specific saved snapshot to a specific owner and makes that link verifiable by anyone, without us being involved.
The key difference from an ordinary cloud backup: a cloud backup has an owner, an invoice and a “terminate service” button. An Arweave record has no owner in that sense. It does not depend on whether we pay for hosting next month, or whether we exist at all.
That is precisely why sixty-five per cent of everything the router receives goes to the treasury to buy AR, the Arweave network token. It is not a reserve for future development; it is a direct payment for the place where your memory will sit after us. A promise of permanence with no storage purchase behind it is just a line in an advert.
The other side of immutability is fairer said up front: what is written to Arweave cannot be erased. Not by us, not by you, not by a court. That is exactly why access to the contents is closed with a key — what gets destroyed is not the record but the ability to read it. There is a separate walkthrough on that below.
The sequence is this. You write a message; it lands in working memory and takes part in the conversation instantly. In parallel, meaningful fragments move to the second layer and become part of semantic search. Once an hour — or immediately, as soon as the dialogue file passes ninety kilobytes — what has accumulated goes to the third layer.
There is not a single button press anywhere in that chain, and there never will be. Manual saving would mean that one day somebody does not press it: tired, distracted, in a hurry. Memory that depends on the owner's discipline is not memory, it is a diary.
The ninety-kilobyte threshold exists for exactly one case — a very long and very important conversation. Waiting out the full hour when twenty minutes have produced a book's worth of text would be unreasonable, so size fires before the clock does.
Both rules apply on every plan, including the free one. Saving your correspondence is the foundation of the system, not a paid feature: charging a person for the right not to lose their own words strikes us as indecent.
Eight actions. The fourth takes longest, because it means waiting until tomorrow.
You register once, on the main site, in the cabinet section. Your email address becomes your permanent identity in the system, and it is where letters about access and charges arrive. Creating a second account is unnecessary and harmful: memory is tied to the account, and a second sign-in tears it into two halves that can never be joined.
Memory starts working from the first line, not after you switch some setting on. Do not try to “upload your biography” straight away — almost everyone does, and it is the worst way to begin. Just talk about what is on your mind today: the semantic layer builds better out of real conversation than out of a questionnaire.
A copy goes to permanent storage automatically: once an hour, or immediately once the dialogue file passes ninety kilobytes. Nothing to press, nothing to confirm. If the conversation was short you will be waiting until the hour is up, which is normal and not a sign of anything being broken.
The single most useful check there is: come back a day later and ask about something from yesterday's conversation without naming dates or exact wording. If the assistant answers to the point, the second layer is working — semantic search, not leftover context sitting in RAM.
The cabinet shows everything that has accumulated: your conversation folder, your plan and the next charge, the people you invited and the income from them. Walk through every figure and make sure you know where the number comes from. A figure whose meaning you do not understand will eventually push you into a bad decision.
Spark at fifteen dollars a month is basic access plus memory. Family Archive at a hundred adds extended limits, personal knowledge bases and family access. Digital DNA is one thousand dollars once per device and two hundred a month after that, for the personal perimeter and permanent recording of identity. Working upwards costs less than working downwards.
Your invitation link sits in the cabinet. Before you send it anywhere, read the level-alignment rule below: referral income is calculated from the amount of your own plan, so an invitee on a more expensive plan will bring you less than they otherwise could. The difference is shown to you directly as lost opportunity.
Deletion here is real: the key is destroyed and the contents stop being readable by anyone, us included. It cannot be recovered for money, on request, or by court order. That is a strong privacy guarantee and, at the same time, a button with no way back — better understood in advance than in a moment of irritation.
Eight screens in order. Every number is explained by what stands behind it.
Your personal folder: everything saved, with timestamps. Check it not for the contents but for the rhythm — entries should appear roughly once an hour on days you talked. A full day's gap during active correspondence is the one thing worth writing to us about.
Your personal link sits here and works immediately — there is no application to file and nothing to approve. Everything that happens through it lands on the neighbouring screens, the grid and the income. Before you send it anywhere, read the level-alignment rule: it decides how much actually gets credited to you.
This screen appears from Family Archive upwards: it holds the materials you want the assistant to lean on when it answers. It is not the same thing as conversation memory — memory assembles itself out of talking, while a knowledge base is filled by you and stays under your control.
Three levels of invitees: the people you invited are L1; the people they invited are L2; the circle after that is L3. The grid goes no deeper than three, and that is not an interface limit but a condition of the smart contract: beyond the third level there are no payouts.
From on-chain memory-usage transactions made by your invitees, you receive fifteen per cent at the first level, seven at the second and three at the third, in $GALATIN. It is a share of a transaction, not of somebody's subscription; confusing those two is the source of half of all disappointment.
A separate line showing the money you did not receive because of the gap between plans. If you are on Spark at fifteen dollars and your invitee is on Family Archive at a hundred, your percentage is calculated from fifteen. The remainder up to the full amount is the lost opportunity. The line exists so that a decision to upgrade is made from a number rather than a feeling.
This shows which tier you are on and what it gives you: limits, personal knowledge bases, family access, the personal perimeter. Upgrading switches on full payouts from your invitees' next payments — nothing is recalculated retroactively.
The last screen is the serious one. It holds what your contents are locked with, and it holds deletion. Re-read the section on keys before you press anything here: there is no way back, and that is deliberate.
The shortest chapter by length and the most expensive by consequence.
The order of operations matters more than the cipher. Contents are locked before they leave for permanent storage, not on arrival. The distinction looks like a formality right up until the first legal request or the first breach: in both cases what sits on the server is an unreadable run of bytes, not text with a note saying “encrypted in transit”.
From that follows the property we consider central to the whole system: we physically cannot read your correspondence. Not “we do not read it as a matter of internal policy”, not “we read it only with your consent” — we cannot, because we do not hold what opens it.
This claim should not be taken on trust, ours included. The right question to ask any service promising privacy is: where does encryption happen, and who holds the key. If the service holds the key, everything else is a question of its good behaviour and its ability to withstand pressure, not of how it is built.
Permanent storage is immutable by nature — that is the entire point of it. A record that can be erased is not permanent, and no amount of promising fixes that. So deletion here is built differently from deletion in an ordinary database.
In an ordinary database, “delete” usually means setting a flag and no longer showing the row. The data stays: in backups, in replication logs, in analytics exports. A year later somebody finds it somewhere nobody was looking.
Here, the key is deleted. The encrypted block stays in the network but turns into noise: without the key it is indistinguishable from random bytes. It cannot be recovered by brute force, by court order or by our goodwill — there is nothing left to recover.
Three consequences worth accepting before rather than after. First: deletion is instant and final, there is no bin. Second: “restore it, I changed my mind” is the one request we cannot fulfil no matter how much we would like to. Third: whoever gets your key gets your memory, and no setting anywhere changes that.
Which gives simple rules of behaviour. Do not delete while angry — come back to that decision tomorrow. Do not hand over access to your account, including to people close to you, unless you want them reading everything. Family Archive exists for shared access, that is what it was built for, and the separation there is done properly.
And separately, the thing we get asked about most. We do not create accounts on your behalf, we do not enter passwords for you, and we never ask you to send a key over any messaging channel under any circumstances. Any email or message asking you to do that in our name is not from us. Our only address is contact@codeofdigitaleternity.com.
The Digital DNA tier — one thousand dollars once per device, then two hundred dollars a month — provides a personal protected perimeter. The difference is not in the number of features but in isolation: your own environment instead of a shared one.
Who genuinely needs it: anyone carrying obligations to third parties. Medical confidentiality, legal privilege, clients' commercial data, any correspondence whose safekeeping you are answerable for to somebody other than yourself. Here isolation is not a luxury but the condition under which you are entitled to use a service like this at all.
Who is overpaying: anyone who wants “the top tier” without being able to say what exactly they are isolating. The privacy guarantee for correspondence does not depend on the plan — it is the same on all of them, free included, because it comes from the key. The perimeter solves a different problem, and buying it for privacy means paying twice for something you already have.
Compared on the rows where people most often get the choice wrong.
| Comparison row | Spark — $15/mo | Family Archive — $100/mo | Digital DNA — $1,000 + $200/mo |
|---|---|---|---|
| Correspondence saved | automatically, once an hour | automatically, once an hour | automatically, once an hour |
| Dialogue size threshold | 90 KB | 90 KB | 90 KB |
| All three PADAM layers | all three | all three | all three |
| Encrypted before storage | yes | yes | yes |
| Usage limits | basic | extended | maximum |
| Personal knowledge bases | no | yes | yes |
| Family access | no | yes | yes |
| Personal protected perimeter | no | no | yes |
| Recording on the blockchain | memory is saved | permanent memory | full recording of identity |
| Base for referral calculation | $15 | $100 | $200 |
| Who it suits | start and find out if you need more | a family and an archive meant to last | obligations to third parties |
Two kinds of node, three levels of depth and one rule that decides everything.
An Ambassador Node is an individual inviting people they know. An Ambassador Team is a company or a partner with an audience of its own. The difference is not the size of the first-level reward but the number of channels: a node has one, a team has three.
Every payout in the system is framed as a Network Validation Fee. The wording was not chosen for elegance: income arises from real memory-usage transactions, not from the act of signing somebody up. Paying for a bare registration is not something the system can do, technically.
This channel works for everyone, regardless of node type. From on-chain memory-usage transactions made by the people you invited, you receive fifteen per cent at the first level, seven at the second and three at the third, paid in $GALATIN.
What is counted is a share of a transaction, not of a subscription. Income is therefore not fixed: it depends on how actively your invitees actually use memory. Someone who registers and disappears brings nothing — that is the design, not an oversight.
The grid goes no deeper than three levels. The smart contract knows no fourth level, and promises of “unlimited depth” from anyone whatsoever simply have no relationship to what the contract says.
The first is the fiat channel, from sales of licences and subscriptions: seven per cent at the first level, three at the second, one at the third. It is calculated from the price of the plans — fifteen, one hundred or two hundred dollars a month respectively.
The second is the buyback channel. If a partner chooses to be paid in $GALATIN, the rates are higher: eight, four and two per cent. The platform then buys the token on the open market for the amount of the reward.
The higher rate on the second channel is neither a gift nor a promotion. Buying from the market creates steady buy pressure, which against a hard cap of ten billion tokens works in favour of every holder at once. The partner is paid more for choosing a form of payout that is useful to the system.
The most misunderstood part of the programme, and simultaneously the most important. Referral income is calculated from the amount of your own plan, not the plan of the person you invited.
Take it in numbers. You are on Spark at fifteen dollars. Someone you invited takes Digital DNA at two hundred dollars a month. Your first-level percentage is calculated from fifteen, not from two hundred. The gap between what was credited and what could have been credited is shown in the cabinet as its own line — that is the lost opportunity.
The rule does not exist to withhold money. It closes off the arrangement where a person on the cheapest plan builds a network of expensive subscribers and lives off somebody else's tier while putting in nothing. Here the level of income is tied to your own participation.
Upgrading switches on full payouts from your invitees' next payments. Nothing is recalculated retroactively: the lost-opportunity line shows the price of waiting, not a debt that will be repaid to you.
The Solana router split in full. It always adds up to one hundred per cent.
| Share | Destination | What it means in practice |
|---|---|---|
| 5% | Founder's Fund | Developing the ecosystem and maintaining what already exists. |
| 5% | Burn | Tokens destroyed irreversibly. Supply shrinks against a hard cap of 10,000,000,000. |
| 15% | Referral L1 | To whoever invited the user directly. |
| 7% | Referral L2 | To whoever invited the inviter. |
| 3% | Referral L3 | The third circle is the last; beyond it there are no payouts. |
| 65% | Treasury | Buying AR for the Arweave pool — a direct payment for permanent storage. |
| Empty level | Burn | If there is no referral at L1, L2 or L3, that share does not stay with the platform: it goes to burn. |
Each of them has cost somebody time, money or part of an archive.
There isn't one and there won't be. Saving happens by itself: once an hour, or immediately past ninety kilobytes. People used to manual backups spend months hunting for the button and conclude that nothing works.
A second email address means a second memory with no link to the first. They cannot be merged back: the contents are locked with different keys. Memory is tied to the account rather than to a device or a domain, which is why one sign-in always beats two.
Deletion destroys the key instantly and finally. There is no bin, no undo, and contacting support does not help — there is physically nothing to restore. The rule is simple: any decision to delete is made the following day.
Referral income is calculated from your plan. Inviting people onto expensive plans while sitting on fifteen dollars means creating your own lost opportunity — and it will be visible in the cabinet as its own line.
Chat history is the first layer and it is short-lived. Permanent memory is the third layer, an immutable record. Between them stands the second, semantic layer, and that is what makes recognition possible. Demanding one layer's properties from another leads nowhere.
Trying to retell your life on day one produces a worse result than a month of ordinary conversation. Semantic memory is built from the context of live speech; a questionnaire yields dry facts without connections, and the connections are exactly what makes recognition possible.
Technical material, but without the jargon.
So you get what you paid for, rather than a tenth of what is available. Half of all support requests are questions already answered inside the cabinet by a figure the person did not know to look at.
Understanding the design before buying is more honest than trusting promises in an advert. What the system cannot do is written down here too: deleted content does not come back, there is no fourth level in the grid, and memory is not a substitute for backing up your work files.
You can only explain to others what you understand yourself. The level-alignment rule and the router split are the two topics people most often get wrong when retelling them, and both are laid out here with numbers.
Doctors, lawyers, consultants. The section on the personal perimeter and the walkthrough of the forty-two Oracle checks were written for you first — because for you, the cost of a mistake is not your own.
If you need company in learning rather than a manual, the mentor with memory is described at aifa.study.
Twelve answers, none of them evasive.
The walkthroughs on this site are open and require no payment. There is no locked section and no ladder of access being sold — the material exists so that people understand what they are using.
What costs money are the service plans themselves: Spark at fifteen dollars a month, Family Archive at a hundred, Digital DNA at one thousand once per device and two hundred a month after that. The walkthroughs explain how they differ, but reading is not tied to paying.
No. Registration and the cabinet run in a browser on the main site, and memory switches itself on with your first message.
A separate device appears only on the Digital DNA tier — the one thousand dollars is for the device itself. It is part of the personal perimeter, not software you install yourself.
Yes, fully and under the same rules: once an hour, or immediately past ninety kilobytes, into a separate personal folder.
This is a position, not an oversight. The right not to lose your own words should not depend on whether you pay. Plans differ by limits, personal knowledge bases, family access and the perimeter — not by whether saving happens.
No. It is the one request we cannot fulfil under any circumstances.
Deletion destroys the key. The encrypted block remains in the network, but without the key it is indistinguishable from random bytes — there is nothing to restore, for us or for anyone else.
That impossibility is the actual content of the privacy promise: a guarantee with an exception for the service owner guarantees nothing.
Whoever holds the key. Contents are encrypted before being sent to permanent storage, so what sits on the servers is an unreadable run of bytes.
We cannot read your correspondence technically — not “we choose not to as a matter of policy”. The converse follows: if an outsider gets into your account, no setting of ours will stop them. Responsibility for the key sits with the owner, and that is the price of building it this way.
History search matches words. The second layer matches meaning: every fragment of experience is stored as a vector, and things close in sense end up close together even when written in completely different words.
The difference shows up immediately in practice. The question “what did we decide about that whole business with the move” contains not a single word from the original conversation — ordinary search finds nothing, semantic search finds the right passage.
No, and that follows from how the second layer is built. What goes into the request is not the whole archive but the few fragments closest in meaning, so the size of what you have accumulated barely affects the cost of any single answer.
Storage grows; the bill for a conversation does not. That is exactly why memory is split into three layers instead of one large archive.
Because that is what the smart contract says: fifteen per cent at the first level, seven at the second, three at the third. There is no fourth level — not in the interface, and not in the contract.
If a level has no referral, that share does not stay with the platform: it is sent to burn. That is the deflationary part of the router — an empty slot works to reduce token supply.
A separate line showing the difference between what you were credited and what you would have been credited if your plan were no lower than your invitee's.
An example: you are on Spark at fifteen dollars, your invitee is on Digital DNA at two hundred a month. Your percentage is calculated from fifteen. The remainder up to the full amount lands in the lost-opportunity line.
The line exists for one purpose: so that the decision to upgrade is made from a number rather than a feeling. Upgrading switches on full payouts from your invitees' next payments; nothing is recalculated retroactively.
The privacy guarantee for correspondence really is identical across all plans, free included — it comes from the key, not from the price.
The perimeter solves a different problem: isolating the environment. It is needed where you are answerable for other people's data — medical confidentiality, legal privilege, clients' commercial information. Buying Digital DNA for correspondence privacy means paying twice for something you already have.
The Oracle runs forty-two checks on a site and points to specific weaknesses against GDPR and OWASP — not general advice, but locations with an explanation of what exactly is wrong there.
A fix costs five hundred dollars as a one-off and takes forty-eight hours. The report is useful on its own terms too: even if somebody else does the fixing, a list of forty-two items gives a clear order of work.
Our only address is contact@codeofdigitaleternity.com. Letters from other addresses, and any request to send a key, a password or a code in our name, are not from us and should not be answered.
We do not create accounts for users, we do not enter passwords for them, and we never ask anyone to forward a key over a messaging channel under any circumstances. That rule has no exceptions, urgent cases included.