An assistant that is yours rather than a company’s, and that keeps what it knows about you to itself.
The agent that belongs to one person.
Each member’s own agent, bound to their identity and working in their context: organizing what they know, what they have decided, what they are doing and what to give the day to. It participates in anonymous federated learning: raw experience stays at its origin, while patterns or model updates may circulate with depersonalized metadata. Deliberate sharing of a person’s knowledge or narrative remains a separate, authorized act.
It is the personal scale of Dk, not the operational runtime, and not Dk Global. Confusing the three is the easiest mistake to make here.
The problem it addresses
An assistant that belongs to a platform ultimately works for the platform. An agent that belongs to the person is the only version of this anyone should trust with a life.
How it works today. Every helpful assistant is owned by a company that also decides what happens to what it learns about you.
What would change. The agent is bound to an identity you own. It can help you examine values, purposes, decisions and consequences; what becomes collective remains a contextual and contestable choice rather than an automatic extraction of personal life.
Why the rest depends on it. If the most personal layer answers to someone else, none of the guarantees above it mean very much.
Where this stands
This is the thinnest of the components, and saying so is more useful than implying otherwise. Internally it exists as an index note and a portfolio reading of it, an interpretation, not a product decision. There is no specification held back: what the agent may keep, what it may send upward and what it must refuse are genuinely unwritten, and the first document about them decides them.
Nothing described here is implemented. This repository exists so that the first document about it has somewhere to live and someone can argue with it in public.
Scope
- One agent per member, bound to UID
- Personal context that does not travel
- Organizing knowledge, decisions and tasks for one person
- Anonymous federated learning through patterns or model updates with depersonalized metadata, while raw personal context stays local
- Deliberate knowledge or narrative represented across scales only when shareable and authorized
- Values, purposes, complementary capacities and the personal E.C.H. learning cycle
- Consent, refusal, contestability and portability of the personal layer
- Relation to Dknowledge, Dk Local and Dk Global
Not in scope
- A released assistant, agent or application.
- Any handling of real personal data.
- The operational runtime, and the global scale of Dk. Both are other things.
How it fits the whole
The scale of Dk that belongs to one person — and the reason the personal layer exists is the same reason the whole system exists: a person is carrying something they never had the conditions to use, and this agent is where those conditions start.
Dk Personal is the mini version of the intelligence — a core of its own, in the spirit of mixture-of-experts but an architecture of its own — and it is connected to the personal Dknowledge of the person: their knowledge, context and history, kept with them. It is bound to UID, the identity the person owns, and authenticated by Living Cryptography. It reads from Dknowledge and participates in two different cross-scale paths. In the federated path, raw personal experience stays local and only patterns or model updates with depersonalized metadata participate; a pattern validated independently in more instances gains progressively stronger weight in global learning. In the deliberate path, the person chooses what knowledge, narrative or contribution to share, and attribution can be preserved. Neither path makes private context common property. The Academy can shape formation to the same profile, so study, real functions and reflection remain connected. Identity preserves continuity and attribution, but neither support for basic needs nor human worth is calculated from one reputation score. The personal agent may advise and represent context; it does not acquire constitutional authority over the member.
Depends on. dk · uid · lc
First functions
These are concrete and unclaimed. Any of them can be opened as an issue and delivered by one person.
- Write one concrete personal use case: who it is for, what hurts, which data it may touch, the expected result and how success is judged. Nothing else here can be settled before this one is.
- Specify the anonymous federated-learning boundary: which updates may leave, how metadata is depersonalized, and how re-identification is tested and prevented.
- Model the distinction between federated patterns and deliberate authorized sharing, including withdrawal, refusal and a useful lesson that remains private.
- Describe one ordinary day it would actually help with.
How to contribute
Read CONTRIBUTING.md
and GOVERNANCE.md in
the organization. In short: open or find an issue, say in the thread that you are taking
it, branch as fn/<issue-number>-<short-name>, and open a pull request against
master. There is no separate review branch.
Participation is voluntary and implies no compensation, employment or future claim.
Sources of truth
- This repository, for what Dk Personal is and is not.
.drayker/component.yml. The machine-readable contract, validated on every pull request.- drayker.org/project/personal/. The same record inside the portal, with the live board.
- drayker.com/project/personal/. The case for it, in plain terms.
Part of Drayker · content under CC BY 4.0